]> Git Repo - linux.git/blame - fs/nfs/pnfs.c
projects / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
block: check bi_size overflow before merge
[linux.git] / fs / nfs / pnfs.c
CommitLineData
85e174ba
RL		 1/*
2 * pNFS functions to call and manage layout drivers.
3 *
4 * Copyright (c) 2002 [year of first publication]
5 * The Regents of the University of Michigan
6 * All Rights Reserved
7 *
8 * Dean Hildebrand <[email protected]>
9 *
10 * Permission is granted to use, copy, create derivative works, and
11 * redistribute this software and such derivative works for any purpose,
12 * so long as the name of the University of Michigan is not used in
13 * any advertising or publicity pertaining to the use or distribution
14 * of this software without specific, written prior authorization. If
15 * the above copyright notice or any other identification of the
16 * University of Michigan is included in any copy of any portion of
17 * this software, then the disclaimer below must also be included.
18 *
19 * This software is provided as is, without representation or warranty
20 * of any kind either express or implied, including without limitation
21 * the implied warranties of merchantability, fitness for a particular
22 * purpose, or noninfringement. The Regents of the University of
23 * Michigan shall not be liable for any damages, including special,
24 * indirect, incidental, or consequential damages, with respect to any
25 * claim arising out of or in connection with the use of the software,
26 * even if it has been or is hereafter advised of the possibility of
27 * such damages.
28 */
29
30#include <linux/nfs_fs.h>
493292dd 31#include <linux/nfs_page.h>
143cb494 32#include <linux/module.h>
ca440c38 33#include <linux/sort.h>
974cec8c 34#include "internal.h"
85e174ba 35#include "pnfs.h"
64419a9b 36#include "iostat.h"
cc668ab3 37#include "nfs4trace.h"
40dd4b7a 38#include "delegation.h"
8733408d 39#include "nfs42.h"
1b146fcf 40#include "nfs4_fs.h"
85e174ba
RL		 41
42#define NFSDBG_FACILITY NFSDBG_PNFS
25c75333 43#define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
85e174ba 44
02c35fca
FI		 45/* Locking:
46 *
47 * pnfs_spinlock:
48 * protects pnfs_modules_tbl.
49 */
50static DEFINE_SPINLOCK(pnfs_spinlock);
51
52/*
53 * pnfs_modules_tbl holds all pnfs modules
54 */
55static LIST_HEAD(pnfs_modules_tbl);
56
13c13a6a 57static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo);
68f74479
TM		 58static void pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
59 struct list_head *free_me,
60 const struct pnfs_layout_range *range,
61 u32 seq);
fe1cf946
TM		 62static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
63 struct list_head *tmp_list);
aa1e0e3a 64
02c35fca
FI		 65/* Return the registered pnfs layout driver module matching given id */
66static struct pnfs_layoutdriver_type *
67find_pnfs_driver_locked(u32 id)
68{
69 struct pnfs_layoutdriver_type *local;
70
71 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
72 if (local->id == id)
73 goto out;
74 local = NULL;
75out:
76 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
77 return local;
78}
79
85e174ba
RL		 80static struct pnfs_layoutdriver_type *
81find_pnfs_driver(u32 id)
82{
02c35fca
FI		 83 struct pnfs_layoutdriver_type *local;
84
85 spin_lock(&pnfs_spinlock);
86 local = find_pnfs_driver_locked(id);
0a9c63fa
TM		 87 if (local != NULL && !try_module_get(local->owner)) {
88 dprintk("%s: Could not grab reference on module\n", __func__);
89 local = NULL;
90 }
02c35fca
FI		 91 spin_unlock(&pnfs_spinlock);
92 return local;
85e174ba
RL		 93}
94
95void
96unset_pnfs_layoutdriver(struct nfs_server *nfss)
97{
738fd0f3
BH		 98 if (nfss->pnfs_curr_ld) {
99 if (nfss->pnfs_curr_ld->clear_layoutdriver)
100 nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
2a4c8994
TM		 101 /* Decrement the MDS count. Purge the deviceid cache if zero */
102 if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count))
103 nfs4_deviceid_purge_client(nfss->nfs_client);
02c35fca 104 module_put(nfss->pnfs_curr_ld->owner);
738fd0f3 105 }
85e174ba
RL		 106 nfss->pnfs_curr_ld = NULL;
107}
108
ca440c38
JL		 109/*
110 * When the server sends a list of layout types, we choose one in the order
111 * given in the list below.
112 *
113 * FIXME: should this list be configurable in some fashion? module param?
114 * mount option? something else?
115 */
116static const u32 ld_prefs[] = {
117 LAYOUT_SCSI,
118 LAYOUT_BLOCK_VOLUME,
119 LAYOUT_OSD2_OBJECTS,
120 LAYOUT_FLEX_FILES,
121 LAYOUT_NFSV4_1_FILES,
122 0
123};
124
125static int
126ld_cmp(const void *e1, const void *e2)
127{
128 u32 ld1 = *((u32 *)e1);
129 u32 ld2 = *((u32 *)e2);
130 int i;
131
132 for (i = 0; ld_prefs[i] != 0; i++) {
133 if (ld1 == ld_prefs[i])
134 return -1;
135
136 if (ld2 == ld_prefs[i])
137 return 1;
138 }
139 return 0;
140}
141
85e174ba
RL		 142/*
143 * Try to set the server's pnfs module to the pnfs layout type specified by id.
144 * Currently only one pNFS layout driver per filesystem is supported.
145 *
3132e49e 146 * @ids array of layout types supported by MDS.
85e174ba
RL		 147 */
148void
738fd0f3 149set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
ca440c38 150 struct nfs_fsinfo *fsinfo)
85e174ba
RL		 151{
152 struct pnfs_layoutdriver_type *ld_type = NULL;
3132e49e 153 u32 id;
ca440c38 154 int i;
85e174ba 155
19274716
AS		 156 if (fsinfo->nlayouttypes == 0)
157 goto out_no_driver;
85e174ba
RL		 158 if (!(server->nfs_client->cl_exchange_flags &
159 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
3132e49e
JL		 160 printk(KERN_ERR "NFS: %s: cl_exchange_flags 0x%x\n",
161 __func__, server->nfs_client->cl_exchange_flags);
85e174ba
RL		 162 goto out_no_driver;
163 }
3132e49e 164
ca440c38
JL		 165 sort(fsinfo->layouttype, fsinfo->nlayouttypes,
166 sizeof(*fsinfo->layouttype), ld_cmp, NULL);
3132e49e 167
ca440c38
JL		 168 for (i = 0; i < fsinfo->nlayouttypes; i++) {
169 id = fsinfo->layouttype[i];
85e174ba 170 ld_type = find_pnfs_driver(id);
ca440c38
JL		 171 if (!ld_type) {
172 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX,
173 id);
174 ld_type = find_pnfs_driver(id);
175 }
176 if (ld_type)
177 break;
85e174ba 178 }
3132e49e
JL		 179
180 if (!ld_type) {
ca440c38 181 dprintk("%s: No pNFS module found!\n", __func__);
3132e49e
JL		 182 goto out_no_driver;
183 }
184
85e174ba 185 server->pnfs_curr_ld = ld_type;
738fd0f3
BH		 186 if (ld_type->set_layoutdriver
187 && ld_type->set_layoutdriver(server, mntfh)) {
a030889a
WAA		 188 printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
189 "driver %u.\n", __func__, id);
738fd0f3
BH		 190 module_put(ld_type->owner);
191 goto out_no_driver;
192 }
2a4c8994
TM		 193 /* Bump the MDS count */
194 atomic_inc(&server->nfs_client->cl_mds_count);
ea8eecdd 195
85e174ba
RL		 196 dprintk("%s: pNFS module for %u set\n", __func__, id);
197 return;
198
199out_no_driver:
200 dprintk("%s: Using NFSv4 I/O\n", __func__);
201 server->pnfs_curr_ld = NULL;
202}
02c35fca
FI		 203
204int
205pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
206{
207 int status = -EINVAL;
208 struct pnfs_layoutdriver_type *tmp;
209
210 if (ld_type->id == 0) {
a030889a 211 printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
02c35fca
FI		 212 return status;
213 }
b1f69b75 214 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
a030889a 215 printk(KERN_ERR "NFS: %s Layout driver must provide "
b1f69b75
AA		 216 "alloc_lseg and free_lseg.\n", __func__);
217 return status;
218 }
02c35fca
FI		 219
220 spin_lock(&pnfs_spinlock);
221 tmp = find_pnfs_driver_locked(ld_type->id);
222 if (!tmp) {
223 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
224 status = 0;
225 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
226 ld_type->name);
227 } else {
a030889a 228 printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
02c35fca
FI		 229 __func__, ld_type->id);
230 }
231 spin_unlock(&pnfs_spinlock);
232
233 return status;
234}
235EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
236
237void
238pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
239{
240 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
241 spin_lock(&pnfs_spinlock);
242 list_del(&ld_type->pnfs_tblid);
243 spin_unlock(&pnfs_spinlock);
244}
245EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
e5e94017 246
b1f69b75
AA		 247/*
248 * pNFS client layout cache
249 */
250
cc6e5340 251/* Need to hold i_lock if caller does not already hold reference */
43f1b3da 252void
70c3bd2b 253pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo)
e5e94017 254{
2b28a7be 255 refcount_inc(&lo->plh_refcount);
e5e94017
BH		 256}
257
636fb9c8
BH		 258static struct pnfs_layout_hdr *
259pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
260{
261 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
57934278 262 return ld->alloc_layout_hdr(ino, gfp_flags);
636fb9c8
BH		 263}
264
265static void
266pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
267{
9c626381
TM		 268 struct nfs_server *server = NFS_SERVER(lo->plh_inode);
269 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
270
271 if (!list_empty(&lo->plh_layouts)) {
272 struct nfs_client *clp = server->nfs_client;
273
274 spin_lock(&clp->cl_lock);
275 list_del_init(&lo->plh_layouts);
276 spin_unlock(&clp->cl_lock);
277 }
a52458b4 278 put_cred(lo->plh_lc_cred);
57934278 279 return ld->free_layout_hdr(lo);
636fb9c8
BH		 280}
281
e5e94017 282static void
6622c3ea 283pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo)
e5e94017 284{
bb346f63 285 struct nfs_inode *nfsi = NFS_I(lo->plh_inode);
cc6e5340 286 dprintk("%s: freeing layout cache %p\n", __func__, lo);
bb346f63
TM		 287 nfsi->layout = NULL;
288 /* Reset MDS Threshold I/O counters */
289 nfsi->write_io = 0;
290 nfsi->read_io = 0;
e5e94017
BH		 291}
292
b1f69b75 293void
70c3bd2b 294pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
974cec8c 295{
9c6376eb 296 struct inode *inode;
cc6e5340 297
9c6376eb
TM		 298 if (!lo)
299 return;
300 inode = lo->plh_inode;
13c13a6a
TM		 301 pnfs_layoutreturn_before_put_layout_hdr(lo);
302
2b28a7be 303 if (refcount_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
566f8737
PT		 304 if (!list_empty(&lo->plh_segs))
305 WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n");
6622c3ea 306 pnfs_detach_layout_hdr(lo);
cc6e5340 307 spin_unlock(&inode->i_lock);
6622c3ea 308 pnfs_free_layout_hdr(lo);
cc6e5340 309 }
974cec8c
AA		 310}
311
4aab9732
TM		 312static void
313pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode,
314 u32 seq)
315{
316 if (lo->plh_return_iomode != 0 && lo->plh_return_iomode != iomode)
317 iomode = IOMODE_ANY;
318 lo->plh_return_iomode = iomode;
319 set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
320 if (seq != 0) {
321 WARN_ON_ONCE(lo->plh_return_seq != 0 && lo->plh_return_seq != seq);
322 lo->plh_return_seq = seq;
323 }
324}
325
ae5a459d
TM		 326static void
327pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr *lo)
328{
5466d214 329 struct pnfs_layout_segment *lseg;
ae5a459d
TM		 330 lo->plh_return_iomode = 0;
331 lo->plh_return_seq = 0;
332 clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
5466d214
TM		 333 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
334 if (!test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
335 continue;
336 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
337 }
ae5a459d
TM		 338}
339
362fb578
TM		 340static void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo)
341{
342 clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
343 clear_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags);
344 smp_mb__after_atomic();
345 wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
346 rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
347}
348
fe1cf946
TM		 349static void
350pnfs_clear_lseg_state(struct pnfs_layout_segment *lseg,
351 struct list_head *free_me)
352{
353 clear_bit(NFS_LSEG_ROC, &lseg->pls_flags);
354 clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
355 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags))
356 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
357 if (test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
358 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
359}
360
7380020e 361/*
30cb3ee2
TM		 362 * Update the seqid of a layout stateid after receiving
363 * NFS4ERR_OLD_STATEID
7380020e 364 */
30cb3ee2 365bool nfs4_layout_refresh_old_stateid(nfs4_stateid *dst,
ecf84026
TM		 366 struct pnfs_layout_range *dst_range,
367 struct inode *inode)
7380020e
TM		 368{
369 struct pnfs_layout_hdr *lo;
c16467dc
TM		 370 struct pnfs_layout_range range = {
371 .iomode = IOMODE_ANY,
372 .offset = 0,
373 .length = NFS4_MAX_UINT64,
374 };
7380020e 375 bool ret = false;
c16467dc
TM		 376 LIST_HEAD(head);
377 int err;
7380020e
TM		 378
379 spin_lock(&inode->i_lock);
380 lo = NFS_I(inode)->layout;
30cb3ee2
TM		 381 if (lo && pnfs_layout_is_valid(lo) &&
382 nfs4_stateid_match_other(dst, &lo->plh_stateid)) {
383 /* Is our call using the most recent seqid? If so, bump it */
384 if (!nfs4_stateid_is_newer(&lo->plh_stateid, dst)) {
385 nfs4_stateid_seqid_inc(dst);
386 ret = true;
387 goto out;
388 }
389 /* Try to update the seqid to the most recent */
c16467dc
TM		 390 err = pnfs_mark_matching_lsegs_return(lo, &head, &range, 0);
391 if (err != -EBUSY) {
392 dst->seqid = lo->plh_stateid.seqid;
ecf84026 393 *dst_range = range;
c16467dc
TM		 394 ret = true;
395 }
7380020e 396 }
30cb3ee2 397out:
7380020e 398 spin_unlock(&inode->i_lock);
c16467dc 399 pnfs_free_lseg_list(&head);
7380020e
TM		 400 return ret;
401}
402
2454dfea
TM		 403/*
404 * Mark a pnfs_layout_hdr and all associated layout segments as invalid
405 *
406 * In order to continue using the pnfs_layout_hdr, a full recovery
407 * is required.
408 * Note that caller must hold inode->i_lock.
409 */
5f46be04 410int
2454dfea
TM		 411pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo,
412 struct list_head *lseg_list)
413{
414 struct pnfs_layout_range range = {
415 .iomode = IOMODE_ANY,
416 .offset = 0,
417 .length = NFS4_MAX_UINT64,
418 };
fe1cf946 419 struct pnfs_layout_segment *lseg, *next;
2454dfea
TM		 420
421 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
fe1cf946
TM		 422 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
423 pnfs_clear_lseg_state(lseg, lseg_list);
5466d214 424 pnfs_clear_layoutreturn_info(lo);
68f74479 425 pnfs_free_returned_lsegs(lo, lseg_list, &range, 0);
362fb578
TM		 426 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) &&
427 !test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
428 pnfs_clear_layoutreturn_waitbit(lo);
fe1cf946 429 return !list_empty(&lo->plh_segs);
2454dfea
TM		 430}
431
b9e028fd
TM		 432static int
433pnfs_iomode_to_fail_bit(u32 iomode)
434{
435 return iomode == IOMODE_RW ?
436 NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
437}
438
439static void
3e621214 440pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
b9e028fd 441{
25c75333 442 lo->plh_retry_timestamp = jiffies;
39e88fcf 443 if (!test_and_set_bit(fail_bit, &lo->plh_flags))
2b28a7be 444 refcount_inc(&lo->plh_refcount);
3e621214
TM		 445}
446
447static void
448pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
449{
450 if (test_and_clear_bit(fail_bit, &lo->plh_flags))
2b28a7be 451 refcount_dec(&lo->plh_refcount);
3e621214
TM		 452}
453
454static void
455pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
456{
457 struct inode *inode = lo->plh_inode;
115ce575
TM		 458 struct pnfs_layout_range range = {
459 .iomode = iomode,
460 .offset = 0,
461 .length = NFS4_MAX_UINT64,
462 };
463 LIST_HEAD(head);
3e621214
TM		 464
465 spin_lock(&inode->i_lock);
466 pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
6d597e17 467 pnfs_mark_matching_lsegs_invalid(lo, &head, &range, 0);
3e621214 468 spin_unlock(&inode->i_lock);
115ce575 469 pnfs_free_lseg_list(&head);
b9e028fd
TM		 470 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
471 iomode == IOMODE_RW ? "RW" : "READ");
472}
473
474static bool
475pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
476{
25c75333 477 unsigned long start, end;
3e621214
TM		 478 int fail_bit = pnfs_iomode_to_fail_bit(iomode);
479
480 if (test_bit(fail_bit, &lo->plh_flags) == 0)
25c75333
TM		 481 return false;
482 end = jiffies;
483 start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
484 if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
485 /* It is time to retry the failed layoutgets */
3e621214 486 pnfs_layout_clear_fail_bit(lo, fail_bit);
25c75333
TM		 487 return false;
488 }
489 return true;
b9e028fd
TM		 490}
491
974cec8c 492static void
119cef97
TM		 493pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg,
494 const struct pnfs_layout_range *range,
495 const nfs4_stateid *stateid)
974cec8c 496{
566052c5 497 INIT_LIST_HEAD(&lseg->pls_list);
a9bae566 498 INIT_LIST_HEAD(&lseg->pls_lc_list);
eba6dd69 499 refcount_set(&lseg->pls_refcount, 1);
4541d16c 500 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
566052c5 501 lseg->pls_layout = lo;
119cef97
TM		 502 lseg->pls_range = *range;
503 lseg->pls_seq = be32_to_cpu(stateid->seqid);
974cec8c
AA		 504}
505
905ca191 506static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
974cec8c 507{
68f74479
TM		 508 if (lseg != NULL) {
509 struct inode *inode = lseg->pls_layout->plh_inode;
510 NFS_SERVER(inode)->pnfs_curr_ld->free_lseg(lseg);
511 }
974cec8c
AA		 512}
513
d684d2ae 514static void
57036a37
TM		 515pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
516 struct pnfs_layout_segment *lseg)
d684d2ae 517{
d20581aa 518 WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
d684d2ae 519 list_del_init(&lseg->pls_list);
8f0d27dc 520 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
2b28a7be 521 refcount_dec(&lo->plh_refcount);
abb3e1c8
TM		 522 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
523 return;
7b650994
TM		 524 if (list_empty(&lo->plh_segs) &&
525 !test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
526 !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
334a8f37
TM		 527 if (atomic_read(&lo->plh_outstanding) == 0)
528 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
173f77e9 529 clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
2d148c7e 530 }
d684d2ae
FI		 531}
532
68f74479
TM		 533static bool
534pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr *lo,
535 struct pnfs_layout_segment *lseg)
536{
537 if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
538 pnfs_layout_is_valid(lo)) {
4aab9732 539 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
68f74479
TM		 540 list_move_tail(&lseg->pls_list, &lo->plh_return_segs);
541 return true;
542 }
543 return false;
544}
545
bae724ef 546void
9369a431 547pnfs_put_lseg(struct pnfs_layout_segment *lseg)
974cec8c 548{
57036a37 549 struct pnfs_layout_hdr *lo;
d684d2ae
FI		 550 struct inode *inode;
551
552 if (!lseg)
553 return;
554
4541d16c 555 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
eba6dd69 556 refcount_read(&lseg->pls_refcount),
4541d16c 557 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
4ef2e4f8 558
57036a37
TM		 559 lo = lseg->pls_layout;
560 inode = lo->plh_inode;
4ef2e4f8 561
eba6dd69 562 if (refcount_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
faa4a54f
TM		 563 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
564 spin_unlock(&inode->i_lock);
565 return;
566 }
8f0d27dc 567 pnfs_get_layout_hdr(lo);
4ef2e4f8 568 pnfs_layout_remove_lseg(lo, lseg);
68f74479
TM		 569 if (pnfs_cache_lseg_for_layoutreturn(lo, lseg))
570 lseg = NULL;
4ef2e4f8
TM		 571 spin_unlock(&inode->i_lock);
572 pnfs_free_lseg(lseg);
573 pnfs_put_layout_hdr(lo);
4541d16c 574 }
4541d16c 575}
9369a431 576EXPORT_SYMBOL_GPL(pnfs_put_lseg);
974cec8c 577
fb3296eb
BH		 578/*
579 * is l2 fully contained in l1?
580 * start1 end1
581 * [----------------------------------)
582 * start2 end2
583 * [----------------)
584 */
3cb2df17 585static bool
7dc0ac70 586pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
3cb2df17 587 const struct pnfs_layout_range *l2)
fb3296eb
BH		 588{
589 u64 start1 = l1->offset;
17822b20 590 u64 end1 = pnfs_end_offset(start1, l1->length);
fb3296eb 591 u64 start2 = l2->offset;
17822b20 592 u64 end2 = pnfs_end_offset(start2, l2->length);
fb3296eb
BH		 593
594 return (start1 <= start2) && (end1 >= end2);
595}
596
24956804
TM		 597static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
598 struct list_head *tmp_list)
599{
eba6dd69 600 if (!refcount_dec_and_test(&lseg->pls_refcount))
24956804
TM		 601 return false;
602 pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
603 list_add(&lseg->pls_list, tmp_list);
604 return true;
605}
606
4541d16c
FI		 607/* Returns 1 if lseg is removed from list, 0 otherwise */
608static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
609 struct list_head *tmp_list)
610{
611 int rv = 0;
612
613 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
614 /* Remove the reference keeping the lseg in the
615 * list. It will now be removed when all
616 * outstanding io is finished.
617 */
d684d2ae 618 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
eba6dd69 619 refcount_read(&lseg->pls_refcount));
24956804 620 if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
d684d2ae 621 rv = 1;
4541d16c
FI		 622 }
623 return rv;
624}
625
6d597e17
JL		 626/*
627 * Compare 2 layout stateid sequence ids, to see which is newer,
628 * taking into account wraparound issues.
629 */
630static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
631{
632 return (s32)(s1 - s2) > 0;
633}
634
e036f464
TM		 635static bool
636pnfs_should_free_range(const struct pnfs_layout_range *lseg_range,
637 const struct pnfs_layout_range *recall_range)
638{
639 return (recall_range->iomode == IOMODE_ANY ||
640 lseg_range->iomode == recall_range->iomode) &&
641 pnfs_lseg_range_intersecting(lseg_range, recall_range);
642}
643
644static bool
645pnfs_match_lseg_recall(const struct pnfs_layout_segment *lseg,
646 const struct pnfs_layout_range *recall_range,
647 u32 seq)
648{
649 if (seq != 0 && pnfs_seqid_is_newer(lseg->pls_seq, seq))
650 return false;
651 if (recall_range == NULL)
652 return true;
653 return pnfs_should_free_range(&lseg->pls_range, recall_range);
654}
655
6d597e17
JL		 656/**
657 * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
658 * @lo: layout header containing the lsegs
659 * @tmp_list: list head where doomed lsegs should go
660 * @recall_range: optional recall range argument to match (may be NULL)
661 * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
662 *
663 * Walk the list of lsegs in the layout header, and tear down any that should
664 * be destroyed. If "recall_range" is specified then the segment must match
665 * that range. If "seq" is non-zero, then only match segments that were handed
666 * out at or before that sequence.
667 *
668 * Returns number of matching invalid lsegs remaining in list after scanning
669 * it and purging them.
4541d16c 670 */
43f1b3da 671int
49a85061 672pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
4541d16c 673 struct list_head *tmp_list,
6d597e17
JL		 674 const struct pnfs_layout_range *recall_range,
675 u32 seq)
974cec8c
AA		 676{
677 struct pnfs_layout_segment *lseg, *next;
71b39854 678 int remaining = 0;
974cec8c
AA		 679
680 dprintk("%s:Begin lo %p\n", __func__, lo);
681
8006bfba 682 if (list_empty(&lo->plh_segs))
38511722 683 return 0;
4541d16c 684 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 685 if (pnfs_match_lseg_recall(lseg, recall_range, seq)) {
b3dce6a2 686 dprintk("%s: freeing lseg %p iomode %d seq %u "
4541d16c 687 "offset %llu length %llu\n", __func__,
6d597e17
JL		 688 lseg, lseg->pls_range.iomode, lseg->pls_seq,
689 lseg->pls_range.offset, lseg->pls_range.length);
71b39854
TM		 690 if (!mark_lseg_invalid(lseg, tmp_list))
691 remaining++;
4541d16c 692 }
71b39854
TM		 693 dprintk("%s:Return %i\n", __func__, remaining);
694 return remaining;
974cec8c
AA		 695}
696
68f74479
TM		 697static void
698pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
699 struct list_head *free_me,
700 const struct pnfs_layout_range *range,
701 u32 seq)
702{
703 struct pnfs_layout_segment *lseg, *next;
704
705 list_for_each_entry_safe(lseg, next, &lo->plh_return_segs, pls_list) {
706 if (pnfs_match_lseg_recall(lseg, range, seq))
707 list_move_tail(&lseg->pls_list, free_me);
708 }
709}
710
f49f9baa 711/* note free_me must contain lsegs from a single layout_hdr */
43f1b3da 712void
4541d16c 713pnfs_free_lseg_list(struct list_head *free_me)
974cec8c 714{
4541d16c 715 struct pnfs_layout_segment *lseg, *tmp;
f49f9baa
FI		 716
717 if (list_empty(free_me))
718 return;
719
4541d16c 720 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
566052c5 721 list_del(&lseg->pls_list);
905ca191 722 pnfs_free_lseg(lseg);
974cec8c
AA		 723 }
724}
725
e5e94017
BH		 726void
727pnfs_destroy_layout(struct nfs_inode *nfsi)
728{
729 struct pnfs_layout_hdr *lo;
974cec8c 730 LIST_HEAD(tmp_list);
e5e94017
BH		 731
732 spin_lock(&nfsi->vfs_inode.i_lock);
733 lo = nfsi->layout;
734 if (lo) {
3e621214 735 pnfs_get_layout_hdr(lo);
2454dfea 736 pnfs_mark_layout_stateid_invalid(lo, &tmp_list);
3e621214
TM		 737 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
738 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
739 spin_unlock(&nfsi->vfs_inode.i_lock);
740 pnfs_free_lseg_list(&tmp_list);
1f18b82c 741 nfs_commit_inode(&nfsi->vfs_inode, 0);
3e621214
TM		 742 pnfs_put_layout_hdr(lo);
743 } else
744 spin_unlock(&nfsi->vfs_inode.i_lock);
974cec8c 745}
041245c8 746EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
974cec8c 747
fd9a8d71
TM		 748static bool
749pnfs_layout_add_bulk_destroy_list(struct inode *inode,
750 struct list_head *layout_list)
974cec8c
AA		 751{
752 struct pnfs_layout_hdr *lo;
fd9a8d71 753 bool ret = false;
974cec8c 754
fd9a8d71
TM		 755 spin_lock(&inode->i_lock);
756 lo = NFS_I(inode)->layout;
757 if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
758 pnfs_get_layout_hdr(lo);
759 list_add(&lo->plh_bulk_destroy, layout_list);
760 ret = true;
761 }
762 spin_unlock(&inode->i_lock);
763 return ret;
764}
765
766/* Caller must hold rcu_read_lock and clp->cl_lock */
767static int
768pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
769 struct nfs_server *server,
770 struct list_head *layout_list)
5085607d
TM		 771 __must_hold(&clp->cl_lock)
772 __must_hold(RCU)
fd9a8d71
TM		 773{
774 struct pnfs_layout_hdr *lo, *next;
775 struct inode *inode;
776
777 list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
5085607d
TM		 778 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags) ||
779 test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) ||
780 !list_empty(&lo->plh_bulk_destroy))
b85f5620 781 continue;
5085607d
TM		 782 /* If the sb is being destroyed, just bail */
783 if (!nfs_sb_active(server->super))
784 break;
fd9a8d71 785 inode = igrab(lo->plh_inode);
5085607d
TM		 786 if (inode != NULL) {
787 list_del_init(&lo->plh_layouts);
788 if (pnfs_layout_add_bulk_destroy_list(inode,
789 layout_list))
790 continue;
791 rcu_read_unlock();
792 spin_unlock(&clp->cl_lock);
793 iput(inode);
794 } else {
795 rcu_read_unlock();
796 spin_unlock(&clp->cl_lock);
797 set_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags);
798 }
799 nfs_sb_deactive(server->super);
fd9a8d71
TM		 800 spin_lock(&clp->cl_lock);
801 rcu_read_lock();
802 return -EAGAIN;
803 }
804 return 0;
805}
806
807static int
808pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
809 bool is_bulk_recall)
810{
811 struct pnfs_layout_hdr *lo;
812 struct inode *inode;
fd9a8d71
TM		 813 LIST_HEAD(lseg_list);
814 int ret = 0;
815
816 while (!list_empty(layout_list)) {
817 lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
818 plh_bulk_destroy);
819 dprintk("%s freeing layout for inode %lu\n", __func__,
820 lo->plh_inode->i_ino);
821 inode = lo->plh_inode;
7c5d1875
CH		 822
823 pnfs_layoutcommit_inode(inode, false);
824
fd9a8d71
TM		 825 spin_lock(&inode->i_lock);
826 list_del_init(&lo->plh_bulk_destroy);
9fd4b9fc
TM		 827 if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) {
828 if (is_bulk_recall)
829 set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
fd9a8d71 830 ret = -EAGAIN;
9fd4b9fc 831 }
fd9a8d71
TM		 832 spin_unlock(&inode->i_lock);
833 pnfs_free_lseg_list(&lseg_list);
b20135d0
TM		 834 /* Free all lsegs that are attached to commit buckets */
835 nfs_commit_inode(inode, 0);
fd9a8d71 836 pnfs_put_layout_hdr(lo);
5085607d 837 nfs_iput_and_deactive(inode);
fd9a8d71
TM		 838 }
839 return ret;
840}
841
842int
843pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
844 struct nfs_fsid *fsid,
845 bool is_recall)
846{
847 struct nfs_server *server;
848 LIST_HEAD(layout_list);
c47abcf8 849
974cec8c 850 spin_lock(&clp->cl_lock);
6382a441 851 rcu_read_lock();
fd9a8d71 852restart:
6382a441 853 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
fd9a8d71
TM		 854 if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
855 continue;
856 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
857 server,
858 &layout_list) != 0)
859 goto restart;
6382a441
WAA		 860 }
861 rcu_read_unlock();
974cec8c
AA		 862 spin_unlock(&clp->cl_lock);
863
fd9a8d71
TM		 864 if (list_empty(&layout_list))
865 return 0;
866 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
867}
868
869int
870pnfs_destroy_layouts_byclid(struct nfs_client *clp,
871 bool is_recall)
872{
873 struct nfs_server *server;
874 LIST_HEAD(layout_list);
875
876 spin_lock(&clp->cl_lock);
877 rcu_read_lock();
878restart:
879 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
880 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
881 server,
882 &layout_list) != 0)
883 goto restart;
974cec8c 884 }
fd9a8d71
TM		 885 rcu_read_unlock();
886 spin_unlock(&clp->cl_lock);
887
888 if (list_empty(&layout_list))
889 return 0;
890 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
891}
892
893/*
894 * Called by the state manger to remove all layouts established under an
895 * expired lease.
896 */
897void
898pnfs_destroy_all_layouts(struct nfs_client *clp)
899{
900 nfs4_deviceid_mark_client_invalid(clp);
901 nfs4_deviceid_purge_client(clp);
902
903 pnfs_destroy_layouts_byclid(clp, false);
e5e94017
BH		 904}
905
fd6002e9 906/* update lo->plh_stateid with new if is more recent */
43f1b3da
FI		 907void
908pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
909 bool update_barrier)
b1f69b75 910{
ecebb80b 911 u32 oldseq, newseq, new_barrier = 0;
b1f69b75 912
2d2f24ad
TM		 913 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
914 newseq = be32_to_cpu(new->seqid);
2a59a041
TM		 915
916 if (!pnfs_layout_is_valid(lo)) {
917 nfs4_stateid_copy(&lo->plh_stateid, new);
918 lo->plh_barrier = newseq;
919 pnfs_clear_layoutreturn_info(lo);
920 clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
921 return;
922 }
923 if (pnfs_seqid_is_newer(newseq, oldseq)) {
f597c537 924 nfs4_stateid_copy(&lo->plh_stateid, new);
ecebb80b
TM		 925 /*
926 * Because of wraparound, we want to keep the barrier
927 * "close" to the current seqids.
928 */
929 new_barrier = newseq - atomic_read(&lo->plh_outstanding);
43f1b3da 930 }
ecebb80b
TM		 931 if (update_barrier)
932 new_barrier = be32_to_cpu(new->seqid);
933 else if (new_barrier == 0)
934 return;
2a59a041 935 if (pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
ecebb80b 936 lo->plh_barrier = new_barrier;
b1f69b75
AA		 937}
938
cf7d63f1 939static bool
19c54aba
TM		 940pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
941 const nfs4_stateid *stateid)
43f1b3da 942{
19c54aba 943 u32 seqid = be32_to_cpu(stateid->seqid);
25a1a621 944
19c54aba
TM		 945 return !pnfs_seqid_is_newer(seqid, lo->plh_barrier);
946}
947
948/* lget is set to 1 if called from inside send_layoutget call chain */
949static bool
e1c06f80 950pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
19c54aba 951{
f7e8917a 952 return lo->plh_block_lgets ||
e1c06f80 953 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
cf7d63f1
FI		 954}
955
5e36e2a9
FI		 956static struct nfs_server *
957pnfs_find_server(struct inode *inode, struct nfs_open_context *ctx)
958{
959 struct nfs_server *server;
960
78746a38 961 if (inode) {
5e36e2a9 962 server = NFS_SERVER(inode);
78746a38 963 } else {
5e36e2a9
FI		 964 struct dentry *parent_dir = dget_parent(ctx->dentry);
965 server = NFS_SERVER(parent_dir->d_inode);
966 dput(parent_dir);
967 }
968 return server;
969}
970
29a8bfe5
TM		 971static void nfs4_free_pages(struct page **pages, size_t size)
972{
973 int i;
974
975 if (!pages)
976 return;
977
978 for (i = 0; i < size; i++) {
979 if (!pages[i])
980 break;
981 __free_page(pages[i]);
982 }
983 kfree(pages);
984}
985
986static struct page **nfs4_alloc_pages(size_t size, gfp_t gfp_flags)
987{
988 struct page **pages;
989 int i;
990
a2791d3a 991 pages = kmalloc_array(size, sizeof(struct page *), gfp_flags);
29a8bfe5
TM		 992 if (!pages) {
993 dprintk("%s: can't alloc array of %zu pages\n", __func__, size);
994 return NULL;
995 }
996
997 for (i = 0; i < size; i++) {
998 pages[i] = alloc_page(gfp_flags);
999 if (!pages[i]) {
1000 dprintk("%s: failed to allocate page\n", __func__);
a2791d3a 1001 nfs4_free_pages(pages, i);
29a8bfe5
TM		 1002 return NULL;
1003 }
1004 }
1005
1006 return pages;
1007}
1008
587f03de 1009static struct nfs4_layoutget *
5e36e2a9 1010pnfs_alloc_init_layoutget_args(struct inode *ino,
e5e94017 1011 struct nfs_open_context *ctx,
2409a976 1012 const nfs4_stateid *stateid,
e144e539 1013 const struct pnfs_layout_range *range,
587f03de 1014 gfp_t gfp_flags)
e5e94017 1015{
5e36e2a9 1016 struct nfs_server *server = pnfs_find_server(ino, ctx);
28ced9a8 1017 size_t max_reply_sz = server->pnfs_curr_ld->max_layoutget_response;
dacb452d 1018 size_t max_pages = max_response_pages(server);
b1f69b75 1019 struct nfs4_layoutget *lgp;
b1f69b75
AA		 1020
1021 dprintk("--> %s\n", __func__);
e5e94017 1022
83026d80
JL		 1023 lgp = kzalloc(sizeof(*lgp), gfp_flags);
1024 if (lgp == NULL)
587f03de 1025 return NULL;
83026d80 1026
28ced9a8
TM		 1027 if (max_reply_sz) {
1028 size_t npages = (max_reply_sz + PAGE_SIZE - 1) >> PAGE_SHIFT;
1029 if (npages < max_pages)
1030 max_pages = npages;
1031 }
1032
dacb452d
FI		 1033 lgp->args.layout.pages = nfs4_alloc_pages(max_pages, gfp_flags);
1034 if (!lgp->args.layout.pages) {
1035 kfree(lgp);
1036 return NULL;
1037 }
1038 lgp->args.layout.pglen = max_pages * PAGE_SIZE;
1039 lgp->res.layoutp = &lgp->args.layout;
1040
d49e0d5b
TM		 1041 /* Don't confuse uninitialised result and success */
1042 lgp->res.status = -NFS4ERR_DELAY;
83026d80
JL		 1043
1044 lgp->args.minlength = PAGE_SIZE;
1045 if (lgp->args.minlength > range->length)
1046 lgp->args.minlength = range->length;
5e36e2a9
FI		 1047 if (ino) {
1048 loff_t i_size = i_size_read(ino);
1049
1050 if (range->iomode == IOMODE_READ) {
1051 if (range->offset >= i_size)
1052 lgp->args.minlength = 0;
1053 else if (i_size - range->offset < lgp->args.minlength)
1054 lgp->args.minlength = i_size - range->offset;
1055 }
d03ab29d 1056 }
83026d80
JL		 1057 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
1058 pnfs_copy_range(&lgp->args.range, range);
1059 lgp->args.type = server->pnfs_curr_ld->id;
1060 lgp->args.inode = ino;
1061 lgp->args.ctx = get_nfs_open_context(ctx);
183d9e7b 1062 nfs4_stateid_copy(&lgp->args.stateid, stateid);
83026d80 1063 lgp->gfp_flags = gfp_flags;
a52458b4 1064 lgp->cred = get_cred(ctx->cred);
587f03de 1065 return lgp;
974cec8c
AA		 1066}
1067
29a8bfe5
TM		 1068void pnfs_layoutget_free(struct nfs4_layoutget *lgp)
1069{
1070 size_t max_pages = lgp->args.layout.pglen / PAGE_SIZE;
1071
1072 nfs4_free_pages(lgp->args.layout.pages, max_pages);
1073 if (lgp->args.inode)
1074 pnfs_put_layout_hdr(NFS_I(lgp->args.inode)->layout);
a52458b4 1075 put_cred(lgp->cred);
29a8bfe5
TM		 1076 put_nfs_open_context(lgp->args.ctx);
1077 kfree(lgp);
1078}
1079
24956804
TM		 1080static void pnfs_clear_layoutcommit(struct inode *inode,
1081 struct list_head *head)
1082{
1083 struct nfs_inode *nfsi = NFS_I(inode);
1084 struct pnfs_layout_segment *lseg, *tmp;
1085
1086 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1087 return;
1088 list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
1089 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1090 continue;
1091 pnfs_lseg_dec_and_remove_zero(lseg, head);
1092 }
1093}
1094
68f74479 1095void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo,
2a974425 1096 const nfs4_stateid *arg_stateid,
68f74479 1097 const struct pnfs_layout_range *range,
68f74479
TM		 1098 const nfs4_stateid *stateid)
1099{
1100 struct inode *inode = lo->plh_inode;
1101 LIST_HEAD(freeme);
1102
1103 spin_lock(&inode->i_lock);
2a974425
TM		 1104 if (!pnfs_layout_is_valid(lo) || !arg_stateid ||
1105 !nfs4_stateid_match_other(&lo->plh_stateid, arg_stateid))
1106 goto out_unlock;
68f74479 1107 if (stateid) {
2a974425
TM		 1108 u32 seq = be32_to_cpu(arg_stateid->seqid);
1109
68f74479
TM		 1110 pnfs_mark_matching_lsegs_invalid(lo, &freeme, range, seq);
1111 pnfs_free_returned_lsegs(lo, &freeme, range, seq);
1112 pnfs_set_layout_stateid(lo, stateid, true);
1113 } else
1114 pnfs_mark_layout_stateid_invalid(lo, &freeme);
2a974425 1115out_unlock:
68f74479
TM		 1116 pnfs_clear_layoutreturn_waitbit(lo);
1117 spin_unlock(&inode->i_lock);
1118 pnfs_free_lseg_list(&freeme);
1119
1120}
1121
13c13a6a 1122static bool
e5fd1904
TM		 1123pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo,
1124 nfs4_stateid *stateid,
1125 enum pnfs_iomode *iomode)
13c13a6a 1126{
bf0291dd
TM		 1127 /* Serialise LAYOUTGET/LAYOUTRETURN */
1128 if (atomic_read(&lo->plh_outstanding) != 0)
1129 return false;
6604b203 1130 if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
13c13a6a 1131 return false;
6604b203 1132 set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
13c13a6a 1133 pnfs_get_layout_hdr(lo);
e5fd1904
TM		 1134 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
1135 if (stateid != NULL) {
1136 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1137 if (lo->plh_return_seq != 0)
1138 stateid->seqid = cpu_to_be32(lo->plh_return_seq);
1139 }
1140 if (iomode != NULL)
1141 *iomode = lo->plh_return_iomode;
1142 pnfs_clear_layoutreturn_info(lo);
1143 return true;
1144 }
1145 if (stateid != NULL)
1146 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1147 if (iomode != NULL)
1148 *iomode = IOMODE_ANY;
13c13a6a
TM		 1149 return true;
1150}
1151
828ed9ec
TM		 1152static void
1153pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args *args,
1154 struct pnfs_layout_hdr *lo,
1155 const nfs4_stateid *stateid,
1156 enum pnfs_iomode iomode)
1157{
1158 struct inode *inode = lo->plh_inode;
1159
1160 args->layout_type = NFS_SERVER(inode)->pnfs_curr_ld->id;
1161 args->inode = inode;
1162 args->range.iomode = iomode;
1163 args->range.offset = 0;
1164 args->range.length = NFS4_MAX_UINT64;
1165 args->layout = lo;
1166 nfs4_stateid_copy(&args->stateid, stateid);
1167}
1168
f40eb5d0 1169static int
ed429d6b 1170pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, const nfs4_stateid *stateid,
6c16605d 1171 enum pnfs_iomode iomode, bool sync)
f40eb5d0
PT		 1172{
1173 struct inode *ino = lo->plh_inode;
287bd3e9 1174 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
f40eb5d0
PT		 1175 struct nfs4_layoutreturn *lrp;
1176 int status = 0;
1177
e4af440a 1178 lrp = kzalloc(sizeof(*lrp), GFP_NOFS);
f40eb5d0
PT		 1179 if (unlikely(lrp == NULL)) {
1180 status = -ENOMEM;
1181 spin_lock(&ino->i_lock);
d67ae825 1182 pnfs_clear_layoutreturn_waitbit(lo);
f40eb5d0
PT		 1183 spin_unlock(&ino->i_lock);
1184 pnfs_put_layout_hdr(lo);
1185 goto out;
1186 }
1187
828ed9ec 1188 pnfs_init_layoutreturn_args(&lrp->args, lo, stateid, iomode);
4d796d75 1189 lrp->args.ld_private = &lrp->ld_private;
f40eb5d0
PT		 1190 lrp->clp = NFS_SERVER(ino)->nfs_client;
1191 lrp->cred = lo->plh_lc_cred;
287bd3e9
TM		 1192 if (ld->prepare_layoutreturn)
1193 ld->prepare_layoutreturn(&lrp->args);
f40eb5d0 1194
6c16605d 1195 status = nfs4_proc_layoutreturn(lrp, sync);
f40eb5d0
PT		 1196out:
1197 dprintk("<-- %s status: %d\n", __func__, status);
1198 return status;
1199}
1200
13c13a6a
TM		 1201/* Return true if layoutreturn is needed */
1202static bool
1203pnfs_layout_need_return(struct pnfs_layout_hdr *lo)
1204{
1205 struct pnfs_layout_segment *s;
f0b42981
TM		 1206 enum pnfs_iomode iomode;
1207 u32 seq;
13c13a6a 1208
2370abda 1209 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a
TM		 1210 return false;
1211
f0b42981
TM		 1212 seq = lo->plh_return_seq;
1213 iomode = lo->plh_return_iomode;
1214
1215 /* Defer layoutreturn until all recalled lsegs are done */
13c13a6a 1216 list_for_each_entry(s, &lo->plh_segs, pls_list) {
f0b42981
TM		 1217 if (seq && pnfs_seqid_is_newer(s->pls_seq, seq))
1218 continue;
1219 if (iomode != IOMODE_ANY && s->pls_range.iomode != iomode)
1220 continue;
13c13a6a
TM		 1221 if (test_bit(NFS_LSEG_LAYOUTRETURN, &s->pls_flags))
1222 return false;
1223 }
1224
1225 return true;
1226}
1227
1228static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo)
1229{
1230 struct inode *inode= lo->plh_inode;
1231
2370abda 1232 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
13c13a6a
TM		 1233 return;
1234 spin_lock(&inode->i_lock);
1235 if (pnfs_layout_need_return(lo)) {
1236 nfs4_stateid stateid;
1237 enum pnfs_iomode iomode;
1238 bool send;
1239
e5fd1904 1240 send = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
13c13a6a
TM		 1241 spin_unlock(&inode->i_lock);
1242 if (send) {
1243 /* Send an async layoutreturn so we dont deadlock */
1244 pnfs_send_layoutreturn(lo, &stateid, iomode, false);
1245 }
1246 } else
1247 spin_unlock(&inode->i_lock);
1248}
1249
293b3b06
AA		 1250/*
1251 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
1252 * when the layout segment list is empty.
1253 *
1254 * Note that a pnfs_layout_hdr can exist with an empty layout segment
1255 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
1256 * deviceid is marked invalid.
1257 */
cbe82603
BH		 1258int
1259_pnfs_return_layout(struct inode *ino)
1260{
1261 struct pnfs_layout_hdr *lo = NULL;
1262 struct nfs_inode *nfsi = NFS_I(ino);
1263 LIST_HEAD(tmp_list);
cbe82603 1264 nfs4_stateid stateid;
24408f52 1265 int status = 0;
93b7f7ad 1266 bool send, valid_layout;
cbe82603 1267
366d5052 1268 dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
cbe82603
BH		 1269
1270 spin_lock(&ino->i_lock);
1271 lo = nfsi->layout;
e5929f3c 1272 if (!lo) {
cbe82603 1273 spin_unlock(&ino->i_lock);
293b3b06
AA		 1274 dprintk("NFS: %s no layout to return\n", __func__);
1275 goto out;
cbe82603 1276 }
cbe82603 1277 /* Reference matched in nfs4_layoutreturn_release */
70c3bd2b 1278 pnfs_get_layout_hdr(lo);
24408f52
TM		 1279 /* Is there an outstanding layoutreturn ? */
1280 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1281 spin_unlock(&ino->i_lock);
1282 if (wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1283 TASK_UNINTERRUPTIBLE))
1284 goto out_put_layout_hdr;
1285 spin_lock(&ino->i_lock);
1286 }
93b7f7ad 1287 valid_layout = pnfs_layout_is_valid(lo);
24956804 1288 pnfs_clear_layoutcommit(ino, &tmp_list);
6d597e17 1289 pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL, 0);
c88953d8
CH		 1290
1291 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
1292 struct pnfs_layout_range range = {
1293 .iomode = IOMODE_ANY,
1294 .offset = 0,
1295 .length = NFS4_MAX_UINT64,
1296 };
1297 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
1298 }
1299
293b3b06 1300 /* Don't send a LAYOUTRETURN if list was initially empty */
93b7f7ad
OK		 1301 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) ||
1302 !valid_layout) {
293b3b06 1303 spin_unlock(&ino->i_lock);
293b3b06 1304 dprintk("NFS: %s no layout segments to return\n", __func__);
7f27392c 1305 goto out_put_layout_hdr;
293b3b06 1306 }
47abadef 1307
e5fd1904 1308 send = pnfs_prepare_layoutreturn(lo, &stateid, NULL);
cbe82603 1309 spin_unlock(&ino->i_lock);
7f27392c 1310 if (send)
ed429d6b 1311 status = pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true);
7f27392c 1312out_put_layout_hdr:
ee6625a9 1313 pnfs_free_lseg_list(&tmp_list);
7f27392c 1314 pnfs_put_layout_hdr(lo);
cbe82603
BH		 1315out:
1316 dprintk("<-- %s status: %d\n", __func__, status);
1317 return status;
1318}
1319
24028672
TM		 1320int
1321pnfs_commit_and_return_layout(struct inode *inode)
1322{
1323 struct pnfs_layout_hdr *lo;
1324 int ret;
1325
1326 spin_lock(&inode->i_lock);
1327 lo = NFS_I(inode)->layout;
1328 if (lo == NULL) {
1329 spin_unlock(&inode->i_lock);
1330 return 0;
1331 }
1332 pnfs_get_layout_hdr(lo);
1333 /* Block new layoutgets and read/write to ds */
1334 lo->plh_block_lgets++;
1335 spin_unlock(&inode->i_lock);
1336 filemap_fdatawait(inode->i_mapping);
1337 ret = pnfs_layoutcommit_inode(inode, true);
1338 if (ret == 0)
1339 ret = _pnfs_return_layout(inode);
1340 spin_lock(&inode->i_lock);
1341 lo->plh_block_lgets--;
1342 spin_unlock(&inode->i_lock);
1343 pnfs_put_layout_hdr(lo);
1344 return ret;
1345}
1346
1c5bd76d
TM		 1347bool pnfs_roc(struct inode *ino,
1348 struct nfs4_layoutreturn_args *args,
1349 struct nfs4_layoutreturn_res *res,
a52458b4 1350 const struct cred *cred)
f7e8917a 1351{
40dd4b7a
TM		 1352 struct nfs_inode *nfsi = NFS_I(ino);
1353 struct nfs_open_context *ctx;
1354 struct nfs4_state *state;
f7e8917a 1355 struct pnfs_layout_hdr *lo;
1c5bd76d 1356 struct pnfs_layout_segment *lseg, *next;
193e3aa2 1357 nfs4_stateid stateid;
1c5bd76d
TM		 1358 enum pnfs_iomode iomode = 0;
1359 bool layoutreturn = false, roc = false;
e71708d4 1360 bool skip_read = false;
f7e8917a 1361
1c5bd76d
TM		 1362 if (!nfs_have_layout(ino))
1363 return false;
29ade5db 1364retry:
0de43976 1365 rcu_read_lock();
f7e8917a 1366 spin_lock(&ino->i_lock);
40dd4b7a 1367 lo = nfsi->layout;
0cdc329e 1368 if (!lo || !pnfs_layout_is_valid(lo) ||
9c6376eb
TM		 1369 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1370 lo = NULL;
40dd4b7a 1371 goto out_noroc;
9c6376eb
TM		 1372 }
1373 pnfs_get_layout_hdr(lo);
29ade5db 1374 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
29ade5db 1375 spin_unlock(&ino->i_lock);
0de43976 1376 rcu_read_unlock();
29ade5db
TM		 1377 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1378 TASK_UNINTERRUPTIBLE);
1379 pnfs_put_layout_hdr(lo);
1380 goto retry;
1381 }
40dd4b7a 1382
e755d638 1383 /* no roc if we hold a delegation */
e71708d4
TM		 1384 if (nfs4_check_delegation(ino, FMODE_READ)) {
1385 if (nfs4_check_delegation(ino, FMODE_WRITE))
1386 goto out_noroc;
1387 skip_read = true;
1388 }
40dd4b7a 1389
0de43976 1390 list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
40dd4b7a 1391 state = ctx->state;
e71708d4
TM		 1392 if (state == NULL)
1393 continue;
40dd4b7a 1394 /* Don't return layout if there is open file state */
e71708d4 1395 if (state->state & FMODE_WRITE)
40dd4b7a 1396 goto out_noroc;
e71708d4
TM		 1397 if (state->state & FMODE_READ)
1398 skip_read = true;
40dd4b7a
TM		 1399 }
1400
e755d638 1401
1c5bd76d 1402 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) {
e71708d4
TM		 1403 if (skip_read && lseg->pls_range.iomode == IOMODE_READ)
1404 continue;
e755d638 1405 /* If we are sending layoutreturn, invalidate all valid lsegs */
1c5bd76d
TM		 1406 if (!test_and_clear_bit(NFS_LSEG_ROC, &lseg->pls_flags))
1407 continue;
1408 /*
1409 * Note: mark lseg for return so pnfs_layout_remove_lseg
1410 * doesn't invalidate the layout for us.
1411 */
1412 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1413 if (!mark_lseg_invalid(lseg, &lo->plh_return_segs))
1414 continue;
1415 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
69820d22
TM		 1416 }
1417
1c5bd76d
TM		 1418 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1419 goto out_noroc;
69820d22 1420
500d701f 1421 /* ROC in two conditions:
e755d638
PT		 1422 * 1. there are ROC lsegs
1423 * 2. we don't send layoutreturn
e755d638 1424 */
1c5bd76d
TM		 1425 /* lo ref dropped in pnfs_roc_release() */
1426 layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
1427 /* If the creds don't match, we can't compound the layoutreturn */
1428 if (!layoutreturn || cred != lo->plh_lc_cred)
1429 goto out_noroc;
1430
1431 roc = layoutreturn;
1432 pnfs_init_layoutreturn_args(args, lo, &stateid, iomode);
1433 res->lrs_present = 0;
1434 layoutreturn = false;
f7e8917a 1435
40dd4b7a 1436out_noroc:
f7e8917a 1437 spin_unlock(&ino->i_lock);
0de43976 1438 rcu_read_unlock();
e755d638 1439 pnfs_layoutcommit_inode(ino, true);
287bd3e9
TM		 1440 if (roc) {
1441 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1442 if (ld->prepare_layoutreturn)
1443 ld->prepare_layoutreturn(args);
9c6376eb 1444 pnfs_put_layout_hdr(lo);
287bd3e9
TM		 1445 return true;
1446 }
e755d638 1447 if (layoutreturn)
1c5bd76d 1448 pnfs_send_layoutreturn(lo, &stateid, iomode, true);
9c6376eb 1449 pnfs_put_layout_hdr(lo);
287bd3e9 1450 return false;
f7e8917a
FI		 1451}
1452
287a9c55
TM		 1453int pnfs_roc_done(struct rpc_task *task, struct inode *inode,
1454 struct nfs4_layoutreturn_args **argpp,
1455 struct nfs4_layoutreturn_res **respp,
1456 int *ret)
1457{
1458 struct nfs4_layoutreturn_args *arg = *argpp;
1459 int retval = -EAGAIN;
1460
1461 if (!arg)
1462 return 0;
1463 /* Handle Layoutreturn errors */
1464 switch (*ret) {
1465 case 0:
1466 retval = 0;
1467 break;
6109bcf7
TM		 1468 case -NFS4ERR_NOMATCHING_LAYOUT:
1469 /* Was there an RPC level error? If not, retry */
1470 if (task->tk_rpc_status == 0)
1471 break;
1472 /* If the call was not sent, let caller handle it */
1473 if (!RPC_WAS_SENT(task))
1474 return 0;
1475 /*
1476 * Otherwise, assume the call succeeded and
1477 * that we need to release the layout
1478 */
1479 *ret = 0;
1480 (*respp)->lrs_present = 0;
1481 retval = 0;
1482 break;
078a432d
TM		 1483 case -NFS4ERR_DELAY:
1484 /* Let the caller handle the retry */
1485 *ret = -NFS4ERR_NOMATCHING_LAYOUT;
1486 return 0;
287a9c55 1487 case -NFS4ERR_OLD_STATEID:
30cb3ee2 1488 if (!nfs4_layout_refresh_old_stateid(&arg->stateid,
287a9c55
TM		 1489 &arg->range, inode))
1490 break;
1491 *ret = -NFS4ERR_NOMATCHING_LAYOUT;
1492 return -EAGAIN;
1493 }
1494 *argpp = NULL;
1495 *respp = NULL;
1496 return retval;
1497}
1498
1c5bd76d
TM		 1499void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
1500 struct nfs4_layoutreturn_res *res,
1501 int ret)
f7e8917a 1502{
1c5bd76d
TM		 1503 struct pnfs_layout_hdr *lo = args->layout;
1504 const nfs4_stateid *arg_stateid = NULL;
1505 const nfs4_stateid *res_stateid = NULL;
287bd3e9 1506 struct nfs4_xdr_opaque_data *ld_private = args->ld_private;
f7e8917a 1507
9c47b18c
TM		 1508 switch (ret) {
1509 case -NFS4ERR_NOMATCHING_LAYOUT:
1510 break;
1511 case 0:
1c5bd76d
TM		 1512 if (res->lrs_present)
1513 res_stateid = &res->stateid;
9c47b18c
TM		 1514 /* Fallthrough */
1515 default:
1516 arg_stateid = &args->stateid;
1c5bd76d
TM		 1517 }
1518 pnfs_layoutreturn_free_lsegs(lo, arg_stateid, &args->range,
1519 res_stateid);
287bd3e9
TM		 1520 if (ld_private && ld_private->ops && ld_private->ops->free)
1521 ld_private->ops->free(ld_private);
1c5bd76d
TM		 1522 pnfs_put_layout_hdr(lo);
1523 trace_nfs4_layoutreturn_on_close(args->inode, 0);
f7e8917a
FI		 1524}
1525
500d701f
PT		 1526bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task)
1527{
1528 struct nfs_inode *nfsi = NFS_I(ino);
1529 struct pnfs_layout_hdr *lo;
1530 bool sleep = false;
1531
1532 /* we might not have grabbed lo reference. so need to check under
1533 * i_lock */
1534 spin_lock(&ino->i_lock);
1535 lo = nfsi->layout;
ee284e35
TM		 1536 if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1537 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
500d701f 1538 sleep = true;
ee284e35 1539 }
500d701f 1540 spin_unlock(&ino->i_lock);
500d701f
PT		 1541 return sleep;
1542}
1543
b1f69b75
AA		 1544/*
1545 * Compare two layout segments for sorting into layout cache.
1546 * We want to preferentially return RW over RO layouts, so ensure those
1547 * are seen first.
1548 */
1549static s64
7dc0ac70 1550pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
3cb2df17 1551 const struct pnfs_layout_range *l2)
b1f69b75 1552{
fb3296eb
BH		 1553 s64 d;
1554
1555 /* high offset > low offset */
1556 d = l1->offset - l2->offset;
1557 if (d)
1558 return d;
1559
1560 /* short length > long length */
1561 d = l2->length - l1->length;
1562 if (d)
1563 return d;
1564
b1f69b75 1565 /* read > read/write */
fb3296eb 1566 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
b1f69b75
AA		 1567}
1568
03772d2f
TM		 1569static bool
1570pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
1571 const struct pnfs_layout_range *l2)
1572{
1573 return pnfs_lseg_range_cmp(l1, l2) > 0;
1574}
1575
1576static bool
1577pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
1578 struct pnfs_layout_segment *old)
1579{
1580 return false;
1581}
1582
1583void
1584pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1585 struct pnfs_layout_segment *lseg,
1586 bool (*is_after)(const struct pnfs_layout_range *,
1587 const struct pnfs_layout_range *),
1588 bool (*do_merge)(struct pnfs_layout_segment *,
1589 struct pnfs_layout_segment *),
1590 struct list_head *free_me)
974cec8c 1591{
03772d2f 1592 struct pnfs_layout_segment *lp, *tmp;
b1f69b75 1593
974cec8c
AA		 1594 dprintk("%s:Begin\n", __func__);
1595
03772d2f
TM		 1596 list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
1597 if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
1598 continue;
1599 if (do_merge(lseg, lp)) {
1600 mark_lseg_invalid(lp, free_me);
1601 continue;
1602 }
1603 if (is_after(&lseg->pls_range, &lp->pls_range))
b1f69b75 1604 continue;
566052c5 1605 list_add_tail(&lseg->pls_list, &lp->pls_list);
b1f69b75
AA		 1606 dprintk("%s: inserted lseg %p "
1607 "iomode %d offset %llu length %llu before "
1608 "lp %p iomode %d offset %llu length %llu\n",
566052c5
FI		 1609 __func__, lseg, lseg->pls_range.iomode,
1610 lseg->pls_range.offset, lseg->pls_range.length,
1611 lp, lp->pls_range.iomode, lp->pls_range.offset,
1612 lp->pls_range.length);
fb3296eb 1613 goto out;
974cec8c 1614 }
fb3296eb
BH		 1615 list_add_tail(&lseg->pls_list, &lo->plh_segs);
1616 dprintk("%s: inserted lseg %p "
1617 "iomode %d offset %llu length %llu at tail\n",
1618 __func__, lseg, lseg->pls_range.iomode,
1619 lseg->pls_range.offset, lseg->pls_range.length);
1620out:
70c3bd2b 1621 pnfs_get_layout_hdr(lo);
974cec8c
AA		 1622
1623 dprintk("%s:Return\n", __func__);
e5e94017 1624}
03772d2f
TM		 1625EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
1626
1627static void
1628pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1629 struct pnfs_layout_segment *lseg,
1630 struct list_head *free_me)
1631{
1632 struct inode *inode = lo->plh_inode;
1633 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
1634
1635 if (ld->add_lseg != NULL)
1636 ld->add_lseg(lo, lseg, free_me);
1637 else
1638 pnfs_generic_layout_insert_lseg(lo, lseg,
1639 pnfs_lseg_range_is_after,
1640 pnfs_lseg_no_merge,
1641 free_me);
1642}
e5e94017
BH		 1643
1644static struct pnfs_layout_hdr *
9fa40758
PT		 1645alloc_init_layout_hdr(struct inode *ino,
1646 struct nfs_open_context *ctx,
1647 gfp_t gfp_flags)
e5e94017
BH		 1648{
1649 struct pnfs_layout_hdr *lo;
1650
636fb9c8 1651 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
e5e94017
BH		 1652 if (!lo)
1653 return NULL;
2b28a7be 1654 refcount_set(&lo->plh_refcount, 1);
b7edfaa1
FI		 1655 INIT_LIST_HEAD(&lo->plh_layouts);
1656 INIT_LIST_HEAD(&lo->plh_segs);
68f74479 1657 INIT_LIST_HEAD(&lo->plh_return_segs);
fd9a8d71 1658 INIT_LIST_HEAD(&lo->plh_bulk_destroy);
b7edfaa1 1659 lo->plh_inode = ino;
a52458b4 1660 lo->plh_lc_cred = get_cred(ctx->cred);
67a3b721 1661 lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID;
e5e94017
BH		 1662 return lo;
1663}
1664
1665static struct pnfs_layout_hdr *
9fa40758
PT		 1666pnfs_find_alloc_layout(struct inode *ino,
1667 struct nfs_open_context *ctx,
1668 gfp_t gfp_flags)
e5241e43
TM		 1669 __releases(&ino->i_lock)
1670 __acquires(&ino->i_lock)
e5e94017
BH		 1671{
1672 struct nfs_inode *nfsi = NFS_I(ino);
1673 struct pnfs_layout_hdr *new = NULL;
1674
1675 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1676
251ec410
TM		 1677 if (nfsi->layout != NULL)
1678 goto out_existing;
e5e94017 1679 spin_unlock(&ino->i_lock);
9fa40758 1680 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
e5e94017
BH		 1681 spin_lock(&ino->i_lock);
1682
251ec410 1683 if (likely(nfsi->layout == NULL)) { /* Won the race? */
e5e94017 1684 nfsi->layout = new;
251ec410 1685 return new;
7175fe90
YN		 1686 } else if (new != NULL)
1687 pnfs_free_layout_hdr(new);
251ec410
TM		 1688out_existing:
1689 pnfs_get_layout_hdr(nfsi->layout);
e5e94017
BH		 1690 return nfsi->layout;
1691}
1692
b1f69b75
AA		 1693/*
1694 * iomode matching rules:
c7d73af2
TH		 1695 * iomode lseg strict match
1696 * iomode
1697 * ----- ----- ------ -----
1698 * ANY READ N/A true
1699 * ANY RW N/A true
1700 * RW READ N/A false
1701 * RW RW N/A true
1702 * READ READ N/A true
1703 * READ RW true false
1704 * READ RW false true
b1f69b75 1705 */
3cb2df17 1706static bool
7dc0ac70 1707pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
c7d73af2
TH		 1708 const struct pnfs_layout_range *range,
1709 bool strict_iomode)
b1f69b75 1710{
fb3296eb
BH		 1711 struct pnfs_layout_range range1;
1712
1713 if ((range->iomode == IOMODE_RW &&
1714 ls_range->iomode != IOMODE_RW) ||
c7d73af2 1715 (range->iomode != ls_range->iomode &&
6089dd0d 1716 strict_iomode) ||
7dc0ac70 1717 !pnfs_lseg_range_intersecting(ls_range, range))
10db5b7a 1718 return false;
fb3296eb
BH		 1719
1720 /* range1 covers only the first byte in the range */
1721 range1 = *range;
1722 range1.length = 1;
7dc0ac70 1723 return pnfs_lseg_range_contained(ls_range, &range1);
b1f69b75
AA		 1724}
1725
1726/*
1727 * lookup range in layout
1728 */
e5e94017 1729static struct pnfs_layout_segment *
fb3296eb 1730pnfs_find_lseg(struct pnfs_layout_hdr *lo,
c7d73af2
TH		 1731 struct pnfs_layout_range *range,
1732 bool strict_iomode)
e5e94017 1733{
b1f69b75
AA		 1734 struct pnfs_layout_segment *lseg, *ret = NULL;
1735
1736 dprintk("%s:Begin\n", __func__);
1737
b7edfaa1 1738 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
4541d16c 1739 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
c7d73af2
TH		 1740 pnfs_lseg_range_match(&lseg->pls_range, range,
1741 strict_iomode)) {
9369a431 1742 ret = pnfs_get_lseg(lseg);
b1f69b75
AA		 1743 break;
1744 }
b1f69b75
AA		 1745 }
1746
1747 dprintk("%s:Return lseg %p ref %d\n",
eba6dd69 1748 __func__, ret, ret ? refcount_read(&ret->pls_refcount) : 0);
b1f69b75 1749 return ret;
e5e94017
BH		 1750}
1751
d23d61c8
AA		 1752/*
1753 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1754 * to the MDS or over pNFS
1755 *
1756 * The nfs_inode read_io and write_io fields are cumulative counters reset
1757 * when there are no layout segments. Note that in pnfs_update_layout iomode
1758 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1759 * WRITE request.
1760 *
1761 * A return of true means use MDS I/O.
1762 *
1763 * From rfc 5661:
1764 * If a file's size is smaller than the file size threshold, data accesses
1765 * SHOULD be sent to the metadata server. If an I/O request has a length that
1766 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1767 * server. If both file size and I/O size are provided, the client SHOULD
1768 * reach or exceed both thresholds before sending its read or write
1769 * requests to the data server.
1770 */
1771static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1772 struct inode *ino, int iomode)
1773{
1774 struct nfs4_threshold *t = ctx->mdsthreshold;
1775 struct nfs_inode *nfsi = NFS_I(ino);
1776 loff_t fsize = i_size_read(ino);
1777 bool size = false, size_set = false, io = false, io_set = false, ret = false;
1778
1779 if (t == NULL)
1780 return ret;
1781
1782 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1783 __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1784
1785 switch (iomode) {
1786 case IOMODE_READ:
1787 if (t->bm & THRESHOLD_RD) {
1788 dprintk("%s fsize %llu\n", __func__, fsize);
1789 size_set = true;
1790 if (fsize < t->rd_sz)
1791 size = true;
1792 }
1793 if (t->bm & THRESHOLD_RD_IO) {
1794 dprintk("%s nfsi->read_io %llu\n", __func__,
1795 nfsi->read_io);
1796 io_set = true;
1797 if (nfsi->read_io < t->rd_io_sz)
1798 io = true;
1799 }
1800 break;
1801 case IOMODE_RW:
1802 if (t->bm & THRESHOLD_WR) {
1803 dprintk("%s fsize %llu\n", __func__, fsize);
1804 size_set = true;
1805 if (fsize < t->wr_sz)
1806 size = true;
1807 }
1808 if (t->bm & THRESHOLD_WR_IO) {
1809 dprintk("%s nfsi->write_io %llu\n", __func__,
1810 nfsi->write_io);
1811 io_set = true;
1812 if (nfsi->write_io < t->wr_io_sz)
1813 io = true;
1814 }
1815 break;
1816 }
1817 if (size_set && io_set) {
1818 if (size && io)
1819 ret = true;
1820 } else if (size || io)
1821 ret = true;
1822
1823 dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1824 return ret;
1825}
1826
d03360aa 1827static int pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
aa8a45ee
PT		 1828{
1829 /*
1830 * send layoutcommit as it can hold up layoutreturn due to lseg
1831 * reference
1832 */
1833 pnfs_layoutcommit_inode(lo->plh_inode, false);
d03360aa 1834 return wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
2e5b29f0 1835 nfs_wait_bit_killable,
d03360aa 1836 TASK_KILLABLE);
aa8a45ee
PT		 1837}
1838
411ae722
TM		 1839static void nfs_layoutget_begin(struct pnfs_layout_hdr *lo)
1840{
1841 atomic_inc(&lo->plh_outstanding);
1842}
1843
1844static void nfs_layoutget_end(struct pnfs_layout_hdr *lo)
1845{
1846 if (atomic_dec_and_test(&lo->plh_outstanding))
1847 wake_up_var(&lo->plh_outstanding);
1848}
1849
d67ae825
TH		 1850static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1851{
1852 unsigned long *bitlock = &lo->plh_flags;
1853
1854 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1855 smp_mb__after_atomic();
1856 wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1857}
1858
78746a38
FI		 1859static void _add_to_server_list(struct pnfs_layout_hdr *lo,
1860 struct nfs_server *server)
1861{
1862 if (list_empty(&lo->plh_layouts)) {
1863 struct nfs_client *clp = server->nfs_client;
1864
1865 /* The lo must be on the clp list if there is any
1866 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1867 */
1868 spin_lock(&clp->cl_lock);
1869 if (list_empty(&lo->plh_layouts))
1870 list_add_tail(&lo->plh_layouts, &server->layouts);
1871 spin_unlock(&clp->cl_lock);
1872 }
1873}
1874
e5e94017
BH		 1875/*
1876 * Layout segment is retreived from the server if not cached.
1877 * The appropriate layout segment is referenced and returned to the caller.
1878 */
7c24d948 1879struct pnfs_layout_segment *
e5e94017
BH		 1880pnfs_update_layout(struct inode *ino,
1881 struct nfs_open_context *ctx,
fb3296eb
BH		 1882 loff_t pos,
1883 u64 count,
a75b9df9 1884 enum pnfs_iomode iomode,
c7d73af2 1885 bool strict_iomode,
a75b9df9 1886 gfp_t gfp_flags)
e5e94017 1887{
fb3296eb
BH		 1888 struct pnfs_layout_range arg = {
1889 .iomode = iomode,
1890 .offset = pos,
1891 .length = count,
1892 };
70d2f7b1 1893 unsigned pg_offset;
6382a441
WAA		 1894 struct nfs_server *server = NFS_SERVER(ino);
1895 struct nfs_client *clp = server->nfs_client;
183d9e7b 1896 struct pnfs_layout_hdr *lo = NULL;
e5e94017 1897 struct pnfs_layout_segment *lseg = NULL;
587f03de 1898 struct nfs4_layoutget *lgp;
183d9e7b
JL		 1899 nfs4_stateid stateid;
1900 long timeout = 0;
66b53f32 1901 unsigned long giveup = jiffies + (clp->cl_lease_time << 1);
30005121 1902 bool first;
e5e94017 1903
9a4bf31d 1904 if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
183d9e7b 1905 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1906 PNFS_UPDATE_LAYOUT_NO_PNFS);
f86bbcf8 1907 goto out;
9a4bf31d 1908 }
d23d61c8 1909
9a4bf31d 1910 if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
183d9e7b 1911 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1912 PNFS_UPDATE_LAYOUT_MDSTHRESH);
f86bbcf8 1913 goto out;
9a4bf31d 1914 }
d23d61c8 1915
9bf87482 1916lookup_again:
d03360aa
TM		 1917 lseg = ERR_PTR(nfs4_client_recover_expired_lease(clp));
1918 if (IS_ERR(lseg))
1919 goto out;
9bf87482 1920 first = false;
e5e94017 1921 spin_lock(&ino->i_lock);
9fa40758 1922 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
830ffb56
TM		 1923 if (lo == NULL) {
1924 spin_unlock(&ino->i_lock);
183d9e7b 1925 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1926 PNFS_UPDATE_LAYOUT_NOMEM);
830ffb56
TM		 1927 goto out;
1928 }
e5e94017 1929
43f1b3da 1930 /* Do we even need to bother with this? */
a59c30ac 1931 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
183d9e7b 1932 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1933 PNFS_UPDATE_LAYOUT_BULK_RECALL);
43f1b3da 1934 dprintk("%s matches recall, use MDS\n", __func__);
e5e94017
BH		 1935 goto out_unlock;
1936 }
1937
1938 /* if LAYOUTGET already failed once we don't try again */
2e5b29f0 1939 if (pnfs_layout_io_test_failed(lo, iomode)) {
183d9e7b 1940 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 1941 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
e5e94017 1942 goto out_unlock;
9a4bf31d 1943 }
e5e94017 1944
411ae722
TM		 1945 /*
1946 * If the layout segment list is empty, but there are outstanding
1947 * layoutget calls, then they might be subject to a layoutrecall.
1948 */
1949 if (list_empty(&lo->plh_segs) &&
1950 atomic_read(&lo->plh_outstanding) != 0) {
1951 spin_unlock(&ino->i_lock);
d03360aa 1952 lseg = ERR_PTR(wait_var_event_killable(&lo->plh_outstanding,
400417b0 1953 !atomic_read(&lo->plh_outstanding)));
58bbeab4 1954 if (IS_ERR(lseg))
411ae722
TM		 1955 goto out_put_layout_hdr;
1956 pnfs_put_layout_hdr(lo);
1957 goto lookup_again;
1958 }
1959
c7d73af2 1960 lseg = pnfs_find_lseg(lo, &arg, strict_iomode);
183d9e7b
JL		 1961 if (lseg) {
1962 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1963 PNFS_UPDATE_LAYOUT_FOUND_CACHED);
1964 goto out_unlock;
1965 }
1966
183d9e7b
JL		 1967 /*
1968 * Choose a stateid for the LAYOUTGET. If we don't have a layout
1969 * stateid, or it has been invalidated, then we must use the open
1970 * stateid.
1971 */
67a3b721 1972 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
d9aba2b4 1973 int status;
183d9e7b
JL		 1974
1975 /*
1976 * The first layoutget for the file. Need to serialize per
9bf87482
PT		 1977 * RFC 5661 Errata 3208.
1978 */
1979 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
1980 &lo->plh_flags)) {
1981 spin_unlock(&ino->i_lock);
d03360aa
TM		 1982 lseg = ERR_PTR(wait_on_bit(&lo->plh_flags,
1983 NFS_LAYOUT_FIRST_LAYOUTGET,
1984 TASK_KILLABLE));
1985 if (IS_ERR(lseg))
1986 goto out_put_layout_hdr;
9bf87482 1987 pnfs_put_layout_hdr(lo);
183d9e7b 1988 dprintk("%s retrying\n", __func__);
9bf87482
PT		 1989 goto lookup_again;
1990 }
183d9e7b
JL		 1991
1992 first = true;
d9aba2b4 1993 status = nfs4_select_rw_stateid(ctx->state,
70d2f7b1 1994 iomode == IOMODE_RW ? FMODE_WRITE : FMODE_READ,
d9aba2b4
TM		 1995 NULL, &stateid, NULL);
1996 if (status != 0) {
731c74dd 1997 lseg = ERR_PTR(status);
70d2f7b1
TM		 1998 trace_pnfs_update_layout(ino, pos, count,
1999 iomode, lo, lseg,
2000 PNFS_UPDATE_LAYOUT_INVALID_OPEN);
d9aba2b4
TM		 2001 if (status != -EAGAIN)
2002 goto out_unlock;
2003 spin_unlock(&ino->i_lock);
2004 nfs4_schedule_stateid_recovery(server, ctx->state);
2005 pnfs_clear_first_layoutget(lo);
2006 pnfs_put_layout_hdr(lo);
2007 goto lookup_again;
70d2f7b1 2008 }
9bf87482 2009 } else {
183d9e7b 2010 nfs4_stateid_copy(&stateid, &lo->plh_stateid);
9bf87482 2011 }
568e8c49 2012
aa8a45ee
PT		 2013 /*
2014 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
2015 * for LAYOUTRETURN even if first is true.
2016 */
8f70f53a 2017 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
aa8a45ee
PT		 2018 spin_unlock(&ino->i_lock);
2019 dprintk("%s wait for layoutreturn\n", __func__);
d03360aa
TM		 2020 lseg = ERR_PTR(pnfs_prepare_to_retry_layoutget(lo));
2021 if (!IS_ERR(lseg)) {
d67ae825
TH		 2022 if (first)
2023 pnfs_clear_first_layoutget(lo);
aa8a45ee
PT		 2024 pnfs_put_layout_hdr(lo);
2025 dprintk("%s retrying\n", __func__);
183d9e7b
JL		 2026 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
2027 lseg, PNFS_UPDATE_LAYOUT_RETRY);
aa8a45ee
PT		 2028 goto lookup_again;
2029 }
183d9e7b 2030 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 2031 PNFS_UPDATE_LAYOUT_RETURN);
aa8a45ee
PT		 2032 goto out_put_layout_hdr;
2033 }
2034
9a4bf31d 2035 if (pnfs_layoutgets_blocked(lo)) {
183d9e7b 2036 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
9a4bf31d 2037 PNFS_UPDATE_LAYOUT_BLOCKED);
cf7d63f1 2038 goto out_unlock;
9a4bf31d 2039 }
411ae722 2040 nfs_layoutget_begin(lo);
f49f9baa 2041 spin_unlock(&ino->i_lock);
30005121 2042
78746a38 2043 _add_to_server_list(lo, server);
e5e94017 2044
09cbfeaf 2045 pg_offset = arg.offset & ~PAGE_MASK;
707ed5fd
BH		 2046 if (pg_offset) {
2047 arg.offset -= pg_offset;
2048 arg.length += pg_offset;
2049 }
7c24d948 2050 if (arg.length != NFS4_MAX_UINT64)
09cbfeaf 2051 arg.length = PAGE_ALIGN(arg.length);
707ed5fd 2052
5e36e2a9 2053 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &stateid, &arg, gfp_flags);
587f03de
FI		 2054 if (!lgp) {
2055 trace_pnfs_update_layout(ino, pos, count, iomode, lo, NULL,
2056 PNFS_UPDATE_LAYOUT_NOMEM);
411ae722 2057 nfs_layoutget_end(lo);
587f03de
FI		 2058 goto out_put_layout_hdr;
2059 }
2060
dacb452d 2061 lseg = nfs4_proc_layoutget(lgp, &timeout);
183d9e7b
JL		 2062 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2063 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
411ae722 2064 nfs_layoutget_end(lo);
83026d80 2065 if (IS_ERR(lseg)) {
183d9e7b 2066 switch(PTR_ERR(lseg)) {
e85d7ee4 2067 case -EBUSY:
183d9e7b
JL		 2068 if (time_after(jiffies, giveup))
2069 lseg = NULL;
66b53f32
TM		 2070 break;
2071 case -ERECALLCONFLICT:
183d9e7b 2072 case -EAGAIN:
56b38a1f 2073 break;
183d9e7b
JL		 2074 default:
2075 if (!nfs_error_is_fatal(PTR_ERR(lseg))) {
2076 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2077 lseg = NULL;
2078 }
56b38a1f
TM		 2079 goto out_put_layout_hdr;
2080 }
2081 if (lseg) {
2082 if (first)
2083 pnfs_clear_first_layoutget(lo);
2084 trace_pnfs_update_layout(ino, pos, count,
2085 iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY);
2086 pnfs_put_layout_hdr(lo);
2087 goto lookup_again;
83026d80
JL		 2088 }
2089 } else {
2090 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2091 }
2092
830ffb56 2093out_put_layout_hdr:
d67ae825
TH		 2094 if (first)
2095 pnfs_clear_first_layoutget(lo);
d5b9216f
TM		 2096 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2097 PNFS_UPDATE_LAYOUT_EXIT);
70c3bd2b 2098 pnfs_put_layout_hdr(lo);
e5e94017 2099out:
f86bbcf8
TM		 2100 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
2101 "(%s, offset: %llu, length: %llu)\n",
2102 __func__, ino->i_sb->s_id,
2103 (unsigned long long)NFS_FILEID(ino),
d600ad1f 2104 IS_ERR_OR_NULL(lseg) ? "not found" : "found",
f86bbcf8
TM		 2105 iomode==IOMODE_RW ? "read/write" : "read-only",
2106 (unsigned long long)pos,
2107 (unsigned long long)count);
e5e94017
BH		 2108 return lseg;
2109out_unlock:
2110 spin_unlock(&ino->i_lock);
830ffb56 2111 goto out_put_layout_hdr;
e5e94017 2112}
7c24d948 2113EXPORT_SYMBOL_GPL(pnfs_update_layout);
b1f69b75 2114
540d9864
TM		 2115static bool
2116pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
2117{
2118 switch (range->iomode) {
2119 case IOMODE_READ:
2120 case IOMODE_RW:
2121 break;
2122 default:
2123 return false;
2124 }
2125 if (range->offset == NFS4_MAX_UINT64)
2126 return false;
2127 if (range->length == 0)
2128 return false;
2129 if (range->length != NFS4_MAX_UINT64 &&
2130 range->length > NFS4_MAX_UINT64 - range->offset)
2131 return false;
2132 return true;
2133}
2134
78746a38
FI		 2135static struct pnfs_layout_hdr *
2136_pnfs_grab_empty_layout(struct inode *ino, struct nfs_open_context *ctx)
2137{
2138 struct pnfs_layout_hdr *lo;
2139
2140 spin_lock(&ino->i_lock);
2141 lo = pnfs_find_alloc_layout(ino, ctx, GFP_KERNEL);
2142 if (!lo)
2143 goto out_unlock;
2144 if (!test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags))
2145 goto out_unlock;
2146 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
2147 goto out_unlock;
2148 if (pnfs_layoutgets_blocked(lo))
2149 goto out_unlock;
2150 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags))
2151 goto out_unlock;
411ae722 2152 nfs_layoutget_begin(lo);
78746a38
FI		 2153 spin_unlock(&ino->i_lock);
2154 _add_to_server_list(lo, NFS_SERVER(ino));
2155 return lo;
2156
2157out_unlock:
2158 spin_unlock(&ino->i_lock);
2159 pnfs_put_layout_hdr(lo);
2160 return NULL;
2161}
2162
2409a976
FI		 2163extern const nfs4_stateid current_stateid;
2164
2165static void _lgopen_prepare_attached(struct nfs4_opendata *data,
2166 struct nfs_open_context *ctx)
2167{
78746a38
FI		 2168 struct inode *ino = data->dentry->d_inode;
2169 struct pnfs_layout_range rng = {
2170 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2171 IOMODE_RW: IOMODE_READ,
2172 .offset = 0,
2173 .length = NFS4_MAX_UINT64,
2174 };
2175 struct nfs4_layoutget *lgp;
2176 struct pnfs_layout_hdr *lo;
2177
64294b08
TM		 2178 /* Heuristic: don't send layoutget if we have cached data */
2179 if (rng.iomode == IOMODE_READ &&
2180 (i_size_read(ino) == 0 || ino->i_mapping->nrpages != 0))
2181 return;
2182
78746a38
FI		 2183 lo = _pnfs_grab_empty_layout(ino, ctx);
2184 if (!lo)
2185 return;
2186 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &current_stateid,
2187 &rng, GFP_KERNEL);
2188 if (!lgp) {
2189 pnfs_clear_first_layoutget(lo);
2190 pnfs_put_layout_hdr(lo);
2191 return;
2192 }
2193 data->lgp = lgp;
2194 data->o_arg.lg_args = &lgp->args;
2195 data->o_res.lg_res = &lgp->res;
2409a976
FI		 2196}
2197
2198static void _lgopen_prepare_floating(struct nfs4_opendata *data,
2199 struct nfs_open_context *ctx)
2200{
2201 struct pnfs_layout_range rng = {
2202 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2203 IOMODE_RW: IOMODE_READ,
2204 .offset = 0,
2205 .length = NFS4_MAX_UINT64,
2206 };
2207 struct nfs4_layoutget *lgp;
2208
2209 lgp = pnfs_alloc_init_layoutget_args(NULL, ctx, &current_stateid,
2210 &rng, GFP_KERNEL);
2211 if (!lgp)
2212 return;
2213 data->lgp = lgp;
2214 data->o_arg.lg_args = &lgp->args;
2215 data->o_res.lg_res = &lgp->res;
2216}
2217
2218void pnfs_lgopen_prepare(struct nfs4_opendata *data,
2219 struct nfs_open_context *ctx)
2220{
2221 struct nfs_server *server = NFS_SERVER(data->dir->d_inode);
2222
2223 if (!(pnfs_enabled_sb(server) &&
2224 server->pnfs_curr_ld->flags & PNFS_LAYOUTGET_ON_OPEN))
2225 return;
2226 /* Could check on max_ops, but currently hardcoded high enough */
6e01260c
FI		 2227 if (!nfs_server_capable(data->dir->d_inode, NFS_CAP_LGOPEN))
2228 return;
2409a976
FI		 2229 if (data->state)
2230 _lgopen_prepare_attached(data, ctx);
2231 else
2232 _lgopen_prepare_floating(data, ctx);
2233}
2234
2235void pnfs_parse_lgopen(struct inode *ino, struct nfs4_layoutget *lgp,
2236 struct nfs_open_context *ctx)
2237{
2238 struct pnfs_layout_hdr *lo;
2239 struct pnfs_layout_segment *lseg;
c49b5209 2240 struct nfs_server *srv = NFS_SERVER(ino);
2409a976
FI		 2241 u32 iomode;
2242
6e01260c 2243 if (!lgp)
2409a976 2244 return;
6e01260c
FI		 2245 dprintk("%s: entered with status %i\n", __func__, lgp->res.status);
2246 if (lgp->res.status) {
2247 switch (lgp->res.status) {
6e01260c 2248 default:
8dc96566
TM		 2249 break;
2250 /*
2251 * Halt lgopen attempts if the server doesn't recognise
2252 * the "current stateid" value, the layout type, or the
2253 * layoutget operation as being valid.
2254 * Also if it complains about too many ops in the compound
2255 * or of the request/reply being too big.
2256 */
2257 case -NFS4ERR_BAD_STATEID:
2258 case -NFS4ERR_NOTSUPP:
2259 case -NFS4ERR_REP_TOO_BIG:
2260 case -NFS4ERR_REP_TOO_BIG_TO_CACHE:
2261 case -NFS4ERR_REQ_TOO_BIG:
2262 case -NFS4ERR_TOO_MANY_OPS:
2263 case -NFS4ERR_UNKNOWN_LAYOUTTYPE:
c49b5209 2264 srv->caps &= ~NFS_CAP_LGOPEN;
6e01260c
FI		 2265 }
2266 return;
2267 }
2409a976 2268 if (!lgp->args.inode) {
78746a38
FI		 2269 lo = _pnfs_grab_empty_layout(ino, ctx);
2270 if (!lo)
2271 return;
2409a976
FI		 2272 lgp->args.inode = ino;
2273 } else
2274 lo = NFS_I(lgp->args.inode)->layout;
2409a976
FI		 2275
2276 lseg = pnfs_layout_process(lgp);
32f1c28f 2277 if (!IS_ERR(lseg)) {
2409a976
FI		 2278 iomode = lgp->args.range.iomode;
2279 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2280 pnfs_put_lseg(lseg);
2281 }
30ae2412
FI		 2282}
2283
2284void nfs4_lgopen_release(struct nfs4_layoutget *lgp)
2285{
2286 if (lgp != NULL) {
2287 struct inode *inode = lgp->args.inode;
2288 if (inode) {
2289 struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
30ae2412 2290 pnfs_clear_first_layoutget(lo);
411ae722 2291 nfs_layoutget_end(lo);
30ae2412
FI		 2292 }
2293 pnfs_layoutget_free(lgp);
2294 }
2409a976
FI		 2295}
2296
a0b0a6e3 2297struct pnfs_layout_segment *
b1f69b75
AA		 2298pnfs_layout_process(struct nfs4_layoutget *lgp)
2299{
2300 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
2301 struct nfs4_layoutget_res *res = &lgp->res;
2302 struct pnfs_layout_segment *lseg;
b7edfaa1 2303 struct inode *ino = lo->plh_inode;
78096cca 2304 LIST_HEAD(free_me);
540d9864
TM		 2305
2306 if (!pnfs_sanity_check_layout_range(&res->range))
1b3c6d07 2307 return ERR_PTR(-EINVAL);
b1f69b75
AA		 2308
2309 /* Inject layout blob into I/O device driver */
a75b9df9 2310 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1b3c6d07 2311 if (IS_ERR_OR_NULL(lseg)) {
b1f69b75 2312 if (!lseg)
1b3c6d07
JL		 2313 lseg = ERR_PTR(-ENOMEM);
2314
2315 dprintk("%s: Could not allocate layout: error %ld\n",
2316 __func__, PTR_ERR(lseg));
2317 return lseg;
b1f69b75
AA		 2318 }
2319
119cef97 2320 pnfs_init_lseg(lo, lseg, &res->range, &res->stateid);
1013df61 2321
b1f69b75 2322 spin_lock(&ino->i_lock);
e1c06f80 2323 if (pnfs_layoutgets_blocked(lo)) {
43f1b3da 2324 dprintk("%s forget reply due to state\n", __func__);
1b3c6d07 2325 goto out_forget;
43f1b3da 2326 }
038d6493 2327
9888d837
TM		 2328 if (!pnfs_layout_is_valid(lo)) {
2329 /* We have a completely new layout */
2330 pnfs_set_layout_stateid(lo, &res->stateid, true);
2331 } else if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
362f7474
CH		 2332 /* existing state ID, make sure the sequence number matches. */
2333 if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
2334 dprintk("%s forget reply due to sequence\n", __func__);
1b3c6d07 2335 goto out_forget;
362f7474
CH		 2336 }
2337 pnfs_set_layout_stateid(lo, &res->stateid, false);
2338 } else {
2339 /*
2340 * We got an entirely new state ID. Mark all segments for the
9888d837 2341 * inode invalid, and retry the layoutget
362f7474 2342 */
d9b61708 2343 pnfs_mark_layout_stateid_invalid(lo, &free_me);
9888d837 2344 goto out_forget;
362f7474 2345 }
038d6493 2346
9369a431 2347 pnfs_get_lseg(lseg);
03772d2f 2348 pnfs_layout_insert_lseg(lo, lseg, &free_me);
8e0acf90 2349
b1f69b75 2350
3976143b 2351 if (res->return_on_close)
f7e8917a 2352 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
f7e8917a 2353
b1f69b75 2354 spin_unlock(&ino->i_lock);
78096cca 2355 pnfs_free_lseg_list(&free_me);
a0b0a6e3 2356 return lseg;
43f1b3da 2357
1b3c6d07 2358out_forget:
43f1b3da
FI		 2359 spin_unlock(&ino->i_lock);
2360 lseg->pls_layout = lo;
2361 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
1b3c6d07 2362 return ERR_PTR(-EAGAIN);
b1f69b75
AA		 2363}
2364
e0b7d420
TM		 2365static int
2366mark_lseg_invalid_or_return(struct pnfs_layout_segment *lseg,
2367 struct list_head *tmp_list)
2368{
2369 if (!mark_lseg_invalid(lseg, tmp_list))
2370 return 0;
2371 pnfs_cache_lseg_for_layoutreturn(lseg->pls_layout, lseg);
2372 return 1;
2373}
2374
2f215968
TM		 2375/**
2376 * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
2377 * @lo: pointer to layout header
2378 * @tmp_list: list header to be used with pnfs_free_lseg_list()
2379 * @return_range: describe layout segment ranges to be returned
e0b7d420 2380 * @seq: stateid seqid to match
2f215968
TM		 2381 *
2382 * This function is mainly intended for use by layoutrecall. It attempts
2383 * to free the layout segment immediately, or else to mark it for return
2384 * as soon as its reference count drops to zero.
e0b7d420
TM		 2385 *
2386 * Returns
2387 * - 0: a layoutreturn needs to be scheduled.
2388 * - EBUSY: there are layout segment that are still in use.
2389 * - ENOENT: there are no layout segments that need to be returned.
2f215968 2390 */
10335556 2391int
016256df
PT		 2392pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
2393 struct list_head *tmp_list,
6d597e17
JL		 2394 const struct pnfs_layout_range *return_range,
2395 u32 seq)
016256df
PT		 2396{
2397 struct pnfs_layout_segment *lseg, *next;
10335556 2398 int remaining = 0;
016256df
PT		 2399
2400 dprintk("%s:Begin lo %p\n", __func__, lo);
2401
fc7ff367 2402 assert_spin_locked(&lo->plh_inode->i_lock);
016256df
PT		 2403
2404 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
e036f464 2405 if (pnfs_match_lseg_recall(lseg, return_range, seq)) {
016256df
PT		 2406 dprintk("%s: marking lseg %p iomode %d "
2407 "offset %llu length %llu\n", __func__,
2408 lseg, lseg->pls_range.iomode,
2409 lseg->pls_range.offset,
2410 lseg->pls_range.length);
e0b7d420 2411 if (mark_lseg_invalid_or_return(lseg, tmp_list))
2f215968
TM		 2412 continue;
2413 remaining++;
016256df 2414 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
016256df 2415 }
6d597e17 2416
e0b7d420 2417 if (remaining) {
6d597e17 2418 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
e0b7d420
TM		 2419 return -EBUSY;
2420 }
6d597e17 2421
e0b7d420
TM		 2422 if (!list_empty(&lo->plh_return_segs)) {
2423 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2424 return 0;
2425 }
2426
2427 return -ENOENT;
016256df
PT		 2428}
2429
2430void pnfs_error_mark_layout_for_return(struct inode *inode,
2431 struct pnfs_layout_segment *lseg)
2432{
2433 struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
016256df
PT		 2434 struct pnfs_layout_range range = {
2435 .iomode = lseg->pls_range.iomode,
2436 .offset = 0,
2437 .length = NFS4_MAX_UINT64,
2438 };
10335556 2439 bool return_now = false;
016256df
PT		 2440
2441 spin_lock(&inode->i_lock);
bdebfccd
TM		 2442 if (!pnfs_layout_is_valid(lo)) {
2443 spin_unlock(&inode->i_lock);
2444 return;
2445 }
2d6cf5ab 2446 pnfs_set_plh_return_info(lo, range.iomode, 0);
016256df
PT		 2447 /*
2448 * mark all matching lsegs so that we are sure to have no live
2449 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
2450 * for how it works.
2451 */
e0b7d420 2452 if (pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, &range, 0) != -EBUSY) {
10335556 2453 nfs4_stateid stateid;
e5fd1904 2454 enum pnfs_iomode iomode;
10335556 2455
e5fd1904 2456 return_now = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
10335556
TM		 2457 spin_unlock(&inode->i_lock);
2458 if (return_now)
2459 pnfs_send_layoutreturn(lo, &stateid, iomode, false);
2460 } else {
2461 spin_unlock(&inode->i_lock);
2462 nfs_commit_inode(inode, 0);
2463 }
016256df
PT		 2464}
2465EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
2466
b3230e80
TM		 2467void
2468pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor *pgio)
2469{
2470 if (pgio->pg_lseg == NULL ||
2471 test_bit(NFS_LSEG_VALID, &pgio->pg_lseg->pls_flags))
2472 return;
2473 pnfs_put_lseg(pgio->pg_lseg);
2474 pgio->pg_lseg = NULL;
2475}
2476EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout);
2477
08cb5b0f
BC		 2478/*
2479 * Check for any intersection between the request and the pgio->pg_lseg,
2480 * and if none, put this pgio->pg_lseg away.
2481 */
2482static void
2483pnfs_generic_pg_check_range(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2484{
2485 if (pgio->pg_lseg && !pnfs_lseg_request_intersecting(pgio->pg_lseg, req)) {
2486 pnfs_put_lseg(pgio->pg_lseg);
2487 pgio->pg_lseg = NULL;
2488 }
2489}
2490
d8007d4d
TM		 2491void
2492pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2493{
1fd937bd
PT		 2494 u64 rd_size = req->wb_bytes;
2495
b3230e80 2496 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2497 pnfs_generic_pg_check_range(pgio, req);
cb5d04bc
PT		 2498 if (pgio->pg_lseg == NULL) {
2499 if (pgio->pg_dreq == NULL)
2500 rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
2501 else
2502 rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
2503
2504 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
9fcd5960 2505 nfs_req_openctx(req),
cb5d04bc
PT		 2506 req_offset(req),
2507 rd_size,
2508 IOMODE_READ,
c7d73af2 2509 false,
cb5d04bc 2510 GFP_KERNEL);
d600ad1f
PT		 2511 if (IS_ERR(pgio->pg_lseg)) {
2512 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2513 pgio->pg_lseg = NULL;
2514 return;
2515 }
cb5d04bc 2516 }
e885de1a
TM		 2517 /* If no lseg, fall back to read through mds */
2518 if (pgio->pg_lseg == NULL)
1f945357 2519 nfs_pageio_reset_read_mds(pgio);
e885de1a 2520
d8007d4d
TM		 2521}
2522EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
2523
2524void
6296556f
PT		 2525pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
2526 struct nfs_page *req, u64 wb_size)
d8007d4d 2527{
b3230e80 2528 pnfs_generic_pg_check_layout(pgio);
08cb5b0f 2529 pnfs_generic_pg_check_range(pgio, req);
d600ad1f 2530 if (pgio->pg_lseg == NULL) {
cb5d04bc 2531 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
9fcd5960 2532 nfs_req_openctx(req),
cb5d04bc
PT		 2533 req_offset(req),
2534 wb_size,
2535 IOMODE_RW,
c7d73af2 2536 false,
2b17d725 2537 GFP_KERNEL);
d600ad1f
PT		 2538 if (IS_ERR(pgio->pg_lseg)) {
2539 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2540 pgio->pg_lseg = NULL;
2541 return;
2542 }
2543 }
e885de1a
TM		 2544 /* If no lseg, fall back to write through mds */
2545 if (pgio->pg_lseg == NULL)
1f945357 2546 nfs_pageio_reset_write_mds(pgio);
d8007d4d
TM		 2547}
2548EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
2549
180bb5ec
WAA		 2550void
2551pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
2552{
2553 if (desc->pg_lseg) {
2554 pnfs_put_lseg(desc->pg_lseg);
2555 desc->pg_lseg = NULL;
2556 }
2557}
2558EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
2559
b4fdac1a
WAA		 2560/*
2561 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
2562 * of bytes (maximum @req->wb_bytes) that can be coalesced.
2563 */
2564size_t
a7d42ddb
WAA		 2565pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
2566 struct nfs_page *prev, struct nfs_page *req)
94ad1c80 2567{
0f9c429e 2568 unsigned int size;
c5e20cb7 2569 u64 seg_end, req_start, seg_left;
0f9c429e
WAA		 2570
2571 size = nfs_generic_pg_test(pgio, prev, req);
0f9c429e
WAA		 2572 if (!size)
2573 return 0;
94ad1c80 2574
19982ba8 2575 /*
c5e20cb7
WAA		 2576 * 'size' contains the number of bytes left in the current page (up
2577 * to the original size asked for in @req->wb_bytes).
2578 *
2579 * Calculate how many bytes are left in the layout segment
2580 * and if there are less bytes than 'size', return that instead.
19982ba8
TM		 2581 *
2582 * Please also note that 'end_offset' is actually the offset of the
2583 * first byte that lies outside the pnfs_layout_range. FIXME?
2584 *
2585 */
19b54848 2586 if (pgio->pg_lseg) {
17822b20 2587 seg_end = pnfs_end_offset(pgio->pg_lseg->pls_range.offset,
c5e20cb7
WAA		 2588 pgio->pg_lseg->pls_range.length);
2589 req_start = req_offset(req);
08cb5b0f 2590
c5e20cb7 2591 /* start of request is past the last byte of this segment */
08cb5b0f 2592 if (req_start >= seg_end)
19b54848 2593 return 0;
c5e20cb7
WAA		 2594
2595 /* adjust 'size' iff there are fewer bytes left in the
2596 * segment than what nfs_generic_pg_test returned */
2597 seg_left = seg_end - req_start;
2598 if (seg_left < size)
2599 size = (unsigned int)seg_left;
19b54848 2600 }
0f9c429e 2601
19b54848 2602 return size;
94ad1c80 2603}
89a58e32 2604EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
94ad1c80 2605
53113ad3 2606int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
e2fecb21
TM		 2607{
2608 struct nfs_pageio_descriptor pgio;
e2fecb21
TM		 2609
2610 /* Resend all requests through the MDS */
53113ad3
WAA		 2611 nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
2612 hdr->completion_ops);
c7070113 2613 set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags);
53113ad3 2614 return nfs_pageio_resend(&pgio, hdr);
e2fecb21 2615}
e7dd79af 2616EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
e2fecb21 2617
d45f60c6 2618static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
1acbbb4e 2619{
cd841605
FI		 2620
2621 dprintk("pnfs write error = %d\n", hdr->pnfs_error);
2622 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2623 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2624 pnfs_return_layout(hdr->inode);
1acbbb4e 2625 }
6c75dc0d 2626 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2627 hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
1acbbb4e
FI		 2628}
2629
d20581aa
BH		 2630/*
2631 * Called by non rpc-based layout drivers
2632 */
d45f60c6 2633void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
44b83799 2634{
f8417b48 2635 if (likely(!hdr->pnfs_error)) {
67af7611
TM		 2636 pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
2637 hdr->mds_offset + hdr->res.count);
d45f60c6 2638 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2639 }
2640 trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
2641 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2642 pnfs_ld_handle_write_error(hdr);
2643 hdr->mds_ops->rpc_release(hdr);
44b83799 2644}
d20581aa 2645EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
44b83799 2646
dce81290
TM		 2647static void
2648pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2649 struct nfs_pgio_header *hdr)
dce81290 2650{
48d635f1 2651 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2652
6c75dc0d 2653 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2654 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
6c75dc0d 2655 nfs_pageio_reset_write_mds(desc);
a7d42ddb 2656 mirror->pg_recoalesce = 1;
6c75dc0d 2657 }
ba4a76f7 2658 hdr->completion_ops->completion(hdr);
dce81290
TM		 2659}
2660
2661static enum pnfs_try_status
d45f60c6 2662pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
dce81290
TM		 2663 const struct rpc_call_ops *call_ops,
2664 struct pnfs_layout_segment *lseg,
2665 int how)
0382b744 2666{
cd841605 2667 struct inode *inode = hdr->inode;
0382b744
AA		 2668 enum pnfs_try_status trypnfs;
2669 struct nfs_server *nfss = NFS_SERVER(inode);
2670
cd841605 2671 hdr->mds_ops = call_ops;
0382b744
AA		 2672
2673 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
d45f60c6
WAA		 2674 inode->i_ino, hdr->args.count, hdr->args.offset, how);
2675 trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
6c75dc0d 2676 if (trypnfs != PNFS_NOT_ATTEMPTED)
0382b744 2677 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
0382b744
AA		 2678 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2679 return trypnfs;
2680}
2681
dce81290 2682static void
7f714720
WAA		 2683pnfs_do_write(struct nfs_pageio_descriptor *desc,
2684 struct nfs_pgio_header *hdr, int how)
dce81290 2685{
dce81290
TM		 2686 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2687 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 2688 enum pnfs_try_status trypnfs;
dce81290 2689
d45f60c6 2690 trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
37f8aa16
TM		 2691 switch (trypnfs) {
2692 case PNFS_NOT_ATTEMPTED:
d45f60c6 2693 pnfs_write_through_mds(desc, hdr);
37f8aa16
TM		 2694 case PNFS_ATTEMPTED:
2695 break;
2696 case PNFS_TRY_AGAIN:
2697 /* cleanup hdr and prepare to redo pnfs */
2698 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2699 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2700 list_splice_init(&hdr->pages, &mirror->pg_list);
2701 mirror->pg_recoalesce = 1;
2702 }
2703 hdr->mds_ops->rpc_release(hdr);
2704 }
dce81290
TM		 2705}
2706
6c75dc0d
FI		 2707static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
2708{
9369a431 2709 pnfs_put_lseg(hdr->lseg);
1e7f3a48 2710 nfs_pgio_header_free(hdr);
6c75dc0d
FI		 2711}
2712
dce81290
TM		 2713int
2714pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
2715{
6c75dc0d 2716 struct nfs_pgio_header *hdr;
dce81290
TM		 2717 int ret;
2718
1e7f3a48
WAA		 2719 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2720 if (!hdr) {
2bff2288
PT		 2721 desc->pg_error = -ENOMEM;
2722 return desc->pg_error;
dce81290 2723 }
6c75dc0d 2724 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
180bb5ec 2725
9369a431 2726 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 2727 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 2728 if (!ret)
7f714720 2729 pnfs_do_write(desc, hdr, desc->pg_ioflags);
a7d42ddb 2730
6c75dc0d 2731 return ret;
dce81290
TM		 2732}
2733EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
2734
53113ad3 2735int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
62e4a769
TM		 2736{
2737 struct nfs_pageio_descriptor pgio;
2738
1acbbb4e 2739 /* Resend all requests through the MDS */
53113ad3
WAA		 2740 nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
2741 return nfs_pageio_resend(&pgio, hdr);
1acbbb4e 2742}
e7dd79af 2743EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
1acbbb4e 2744
d45f60c6 2745static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
1acbbb4e 2746{
cd841605
FI		 2747 dprintk("pnfs read error = %d\n", hdr->pnfs_error);
2748 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1acbbb4e 2749 PNFS_LAYOUTRET_ON_ERROR) {
cd841605 2750 pnfs_return_layout(hdr->inode);
1acbbb4e 2751 }
4db6e0b7 2752 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
53113ad3 2753 hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
62e4a769
TM		 2754}
2755
d20581aa
BH		 2756/*
2757 * Called by non rpc-based layout drivers
2758 */
d45f60c6 2759void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
d20581aa 2760{
bfc505de 2761 if (likely(!hdr->pnfs_error))
d45f60c6 2762 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
f8417b48
KM		 2763 trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
2764 if (unlikely(hdr->pnfs_error))
d45f60c6
WAA		 2765 pnfs_ld_handle_read_error(hdr);
2766 hdr->mds_ops->rpc_release(hdr);
d20581aa
BH		 2767}
2768EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
2769
493292dd
TM		 2770static void
2771pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
d45f60c6 2772 struct nfs_pgio_header *hdr)
493292dd 2773{
48d635f1 2774 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
a7d42ddb 2775
4db6e0b7 2776 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
a7d42ddb 2777 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
4db6e0b7 2778 nfs_pageio_reset_read_mds(desc);
a7d42ddb 2779 mirror->pg_recoalesce = 1;
4db6e0b7 2780 }
ba4a76f7 2781 hdr->completion_ops->completion(hdr);
493292dd
TM		 2782}
2783
64419a9b
AA		 2784/*
2785 * Call the appropriate parallel I/O subsystem read function.
2786 */
493292dd 2787static enum pnfs_try_status
d45f60c6 2788pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
493292dd
TM		 2789 const struct rpc_call_ops *call_ops,
2790 struct pnfs_layout_segment *lseg)
64419a9b 2791{
cd841605 2792 struct inode *inode = hdr->inode;
64419a9b
AA		 2793 struct nfs_server *nfss = NFS_SERVER(inode);
2794 enum pnfs_try_status trypnfs;
2795
cd841605 2796 hdr->mds_ops = call_ops;
64419a9b
AA		 2797
2798 dprintk("%s: Reading ino:%lu %u@%llu\n",
d45f60c6 2799 __func__, inode->i_ino, hdr->args.count, hdr->args.offset);
64419a9b 2800
d45f60c6 2801 trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
4db6e0b7 2802 if (trypnfs != PNFS_NOT_ATTEMPTED)
64419a9b 2803 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
64419a9b
AA		 2804 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2805 return trypnfs;
2806}
863a3c6c 2807
ceb11e13 2808/* Resend all requests through pnfs. */
1b1bc66b 2809void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr)
ceb11e13
PT		 2810{
2811 struct nfs_pageio_descriptor pgio;
2812
1b1bc66b 2813 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
54e4a0df
TM		 2814 /* Prevent deadlocks with layoutreturn! */
2815 pnfs_put_lseg(hdr->lseg);
2816 hdr->lseg = NULL;
2817
1b1bc66b
WAA		 2818 nfs_pageio_init_read(&pgio, hdr->inode, false,
2819 hdr->completion_ops);
2820 hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr);
2821 }
ceb11e13
PT		 2822}
2823EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
2824
493292dd 2825static void
7f714720 2826pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
493292dd 2827{
493292dd
TM		 2828 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2829 struct pnfs_layout_segment *lseg = desc->pg_lseg;
7f714720 2830 enum pnfs_try_status trypnfs;
493292dd 2831
d45f60c6 2832 trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
6aeafd05
TM		 2833 switch (trypnfs) {
2834 case PNFS_NOT_ATTEMPTED:
d45f60c6 2835 pnfs_read_through_mds(desc, hdr);
6aeafd05
TM		 2836 case PNFS_ATTEMPTED:
2837 break;
2838 case PNFS_TRY_AGAIN:
2839 /* cleanup hdr and prepare to redo pnfs */
2840 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2841 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2842 list_splice_init(&hdr->pages, &mirror->pg_list);
2843 mirror->pg_recoalesce = 1;
2844 }
2845 hdr->mds_ops->rpc_release(hdr);
2846 }
493292dd
TM		 2847}
2848
4db6e0b7
FI		 2849static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
2850{
9369a431 2851 pnfs_put_lseg(hdr->lseg);
1e7f3a48 2852 nfs_pgio_header_free(hdr);
4db6e0b7
FI		 2853}
2854
493292dd
TM		 2855int
2856pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
2857{
4db6e0b7 2858 struct nfs_pgio_header *hdr;
493292dd
TM		 2859 int ret;
2860
1e7f3a48
WAA		 2861 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2862 if (!hdr) {
2bff2288
PT		 2863 desc->pg_error = -ENOMEM;
2864 return desc->pg_error;
493292dd 2865 }
4db6e0b7 2866 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
9369a431 2867 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
ef2c488c 2868 ret = nfs_generic_pgio(desc, hdr);
180bb5ec 2869 if (!ret)
7f714720 2870 pnfs_do_read(desc, hdr);
4db6e0b7 2871 return ret;
493292dd
TM		 2872}
2873EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
2874
71244d9b
TM		 2875static void pnfs_clear_layoutcommitting(struct inode *inode)
2876{
2877 unsigned long *bitlock = &NFS_I(inode)->flags;
2878
2879 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
4e857c58 2880 smp_mb__after_atomic();
71244d9b
TM		 2881 wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
2882}
2883
863a3c6c 2884/*
a9bae566 2885 * There can be multiple RW segments.
863a3c6c 2886 */
a9bae566 2887static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
863a3c6c 2888{
a9bae566 2889 struct pnfs_layout_segment *lseg;
863a3c6c 2890
a9bae566
PT		 2891 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
2892 if (lseg->pls_range.iomode == IOMODE_RW &&
a073dbff 2893 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
a9bae566
PT		 2894 list_add(&lseg->pls_lc_list, listp);
2895 }
863a3c6c
AA		 2896}
2897
a073dbff
TM		 2898static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
2899{
2900 struct pnfs_layout_segment *lseg, *tmp;
a073dbff
TM		 2901
2902 /* Matched by references in pnfs_set_layoutcommit */
2903 list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
2904 list_del_init(&lseg->pls_lc_list);
2905 pnfs_put_lseg(lseg);
2906 }
2907
71244d9b 2908 pnfs_clear_layoutcommitting(inode);
a073dbff
TM		 2909}
2910
1b0ae068
PT		 2911void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
2912{
b9e028fd 2913 pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
1b0ae068
PT		 2914}
2915EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
2916
863a3c6c 2917void
67af7611
TM		 2918pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
2919 loff_t end_pos)
863a3c6c 2920{
cd841605 2921 struct nfs_inode *nfsi = NFS_I(inode);
79a48a1f 2922 bool mark_as_dirty = false;
863a3c6c 2923
cd841605 2924 spin_lock(&inode->i_lock);
863a3c6c 2925 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
29559b11 2926 nfsi->layout->plh_lwb = end_pos;
79a48a1f 2927 mark_as_dirty = true;
863a3c6c 2928 dprintk("%s: Set layoutcommit for inode %lu ",
cd841605 2929 __func__, inode->i_ino);
29559b11
TM		 2930 } else if (end_pos > nfsi->layout->plh_lwb)
2931 nfsi->layout->plh_lwb = end_pos;
67af7611 2932 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
a9bae566 2933 /* references matched in nfs4_layoutcommit_release */
67af7611 2934 pnfs_get_lseg(lseg);
a9bae566 2935 }
cd841605 2936 spin_unlock(&inode->i_lock);
acff5880 2937 dprintk("%s: lseg %p end_pos %llu\n",
67af7611 2938 __func__, lseg, nfsi->layout->plh_lwb);
79a48a1f
WAA		 2939
2940 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
2941 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
2942 if (mark_as_dirty)
cd841605 2943 mark_inode_dirty_sync(inode);
863a3c6c
AA		 2944}
2945EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
2946
db29c089
AA		 2947void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
2948{
2949 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
2950
2951 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
2952 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
a073dbff 2953 pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
db29c089
AA		 2954}
2955
de4b15c7
AA		 2956/*
2957 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
2958 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
2959 * data to disk to allow the server to recover the data if it crashes.
2960 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
2961 * is off, and a COMMIT is sent to a data server, or
2962 * if WRITEs to a data server return NFS_DATA_SYNC.
2963 */
863a3c6c 2964int
ef311537 2965pnfs_layoutcommit_inode(struct inode *inode, bool sync)
863a3c6c 2966{
5f919c9f 2967 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
863a3c6c
AA		 2968 struct nfs4_layoutcommit_data *data;
2969 struct nfs_inode *nfsi = NFS_I(inode);
863a3c6c 2970 loff_t end_pos;
71244d9b 2971 int status;
863a3c6c 2972
71244d9b 2973 if (!pnfs_layoutcommit_outstanding(inode))
de4b15c7
AA		 2974 return 0;
2975
71244d9b 2976 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
92407e75 2977
71244d9b 2978 status = -EAGAIN;
92407e75 2979 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
71244d9b
TM		 2980 if (!sync)
2981 goto out;
74316201 2982 status = wait_on_bit_lock_action(&nfsi->flags,
71244d9b
TM		 2983 NFS_INO_LAYOUTCOMMITTING,
2984 nfs_wait_bit_killable,
2985 TASK_KILLABLE);
92407e75 2986 if (status)
71244d9b 2987 goto out;
92407e75
PT		 2988 }
2989
71244d9b
TM		 2990 status = -ENOMEM;
2991 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
2992 data = kzalloc(sizeof(*data), GFP_NOFS);
2993 if (!data)
2994 goto clear_layoutcommitting;
2995
2996 status = 0;
de4b15c7 2997 spin_lock(&inode->i_lock);
71244d9b
TM		 2998 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
2999 goto out_unlock;
a9bae566 3000
71244d9b 3001 INIT_LIST_HEAD(&data->lseg_list);
a9bae566 3002 pnfs_list_write_lseg(inode, &data->lseg_list);
863a3c6c 3003
acff5880 3004 end_pos = nfsi->layout->plh_lwb;
863a3c6c 3005
f597c537 3006 nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
863a3c6c
AA		 3007 spin_unlock(&inode->i_lock);
3008
3009 data->args.inode = inode;
a52458b4 3010 data->cred = get_cred(nfsi->layout->plh_lc_cred);
863a3c6c
AA		 3011 nfs_fattr_init(&data->fattr);
3012 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
3013 data->res.fattr = &data->fattr;
2e18d4d8
TM		 3014 if (end_pos != 0)
3015 data->args.lastbytewritten = end_pos - 1;
3016 else
3017 data->args.lastbytewritten = U64_MAX;
863a3c6c
AA		 3018 data->res.server = NFS_SERVER(inode);
3019
5f919c9f
CH		 3020 if (ld->prepare_layoutcommit) {
3021 status = ld->prepare_layoutcommit(&data->args);
3022 if (status) {
a52458b4 3023 put_cred(data->cred);
5f919c9f 3024 spin_lock(&inode->i_lock);
29559b11
TM		 3025 set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
3026 if (end_pos > nfsi->layout->plh_lwb)
5f919c9f 3027 nfsi->layout->plh_lwb = end_pos;
3471648a 3028 goto out_unlock;
5f919c9f
CH		 3029 }
3030 }
3031
3032
863a3c6c
AA		 3033 status = nfs4_proc_layoutcommit(data, sync);
3034out:
92407e75
PT		 3035 if (status)
3036 mark_inode_dirty_sync(inode);
863a3c6c
AA		 3037 dprintk("<-- %s status %d\n", __func__, status);
3038 return status;
71244d9b
TM		 3039out_unlock:
3040 spin_unlock(&inode->i_lock);
92407e75 3041 kfree(data);
71244d9b
TM		 3042clear_layoutcommitting:
3043 pnfs_clear_layoutcommitting(inode);
92407e75 3044 goto out;
863a3c6c 3045}
72cff449 3046EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
82be417a 3047
5bb89b47
TM		 3048int
3049pnfs_generic_sync(struct inode *inode, bool datasync)
3050{
3051 return pnfs_layoutcommit_inode(inode, true);
3052}
3053EXPORT_SYMBOL_GPL(pnfs_generic_sync);
3054
82be417a
AA		 3055struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
3056{
3057 struct nfs4_threshold *thp;
3058
3059 thp = kzalloc(sizeof(*thp), GFP_NOFS);
3060 if (!thp) {
3061 dprintk("%s mdsthreshold allocation failed\n", __func__);
3062 return NULL;
3063 }
3064 return thp;
3065}
8733408d 3066
865a7ecb 3067#if IS_ENABLED(CONFIG_NFS_V4_2)
8733408d 3068int
c8ad8894 3069pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
8733408d
PT		 3070{
3071 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
3072 struct nfs_server *server = NFS_SERVER(inode);
1bfe3b25 3073 struct nfs_inode *nfsi = NFS_I(inode);
8733408d
PT		 3074 struct nfs42_layoutstat_data *data;
3075 struct pnfs_layout_hdr *hdr;
3076 int status = 0;
3077
3078 if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
3079 goto out;
3080
6c5a0d89
TM		 3081 if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
3082 goto out;
3083
1bfe3b25
PT		 3084 if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
3085 goto out;
3086
8733408d
PT		 3087 spin_lock(&inode->i_lock);
3088 if (!NFS_I(inode)->layout) {
3089 spin_unlock(&inode->i_lock);
f538d0ba 3090 goto out_clear_layoutstats;
8733408d
PT		 3091 }
3092 hdr = NFS_I(inode)->layout;
3093 pnfs_get_layout_hdr(hdr);
3094 spin_unlock(&inode->i_lock);
3095
c8ad8894 3096 data = kzalloc(sizeof(*data), gfp_flags);
8733408d
PT		 3097 if (!data) {
3098 status = -ENOMEM;
3099 goto out_put;
3100 }
3101
3102 data->args.fh = NFS_FH(inode);
3103 data->args.inode = inode;
8733408d
PT		 3104 status = ld->prepare_layoutstats(&data->args);
3105 if (status)
3106 goto out_free;
3107
3108 status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);
3109
3110out:
3111 dprintk("%s returns %d\n", __func__, status);
3112 return status;
3113
3114out_free:
3115 kfree(data);
3116out_put:
3117 pnfs_put_layout_hdr(hdr);
f538d0ba 3118out_clear_layoutstats:
1bfe3b25
PT		 3119 smp_mb__before_atomic();
3120 clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
3121 smp_mb__after_atomic();
8733408d
PT		 3122 goto out;
3123}
3124EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
865a7ecb 3125#endif
bbf58bf3
TM		 3126
3127unsigned int layoutstats_timer;
3128module_param(layoutstats_timer, uint, 0644);
3129EXPORT_SYMBOL_GPL(layoutstats_timer);
Empty description
This page took 1.023766 seconds and 4 git commands to generate.